1 // SPDX-License-Identifier: (GPL-2.0-only OR BSD-2-Clause)
2 /* Copyright (C) 2015-2018 Netronome Systems, Inc. */
3 
4 /*
5  * nfp_net_common.c
6  * Netronome network device driver: Common functions between PF and VF
7  * Authors: Jakub Kicinski <jakub.kicinski@netronome.com>
8  *          Jason McMullan <jason.mcmullan@netronome.com>
9  *          Rolf Neugebauer <rolf.neugebauer@netronome.com>
10  *          Brad Petrus <brad.petrus@netronome.com>
11  *          Chris Telfer <chris.telfer@netronome.com>
12  */
13 
14 #include <linux/bitfield.h>
15 #include <linux/bpf.h>
16 #include <linux/bpf_trace.h>
17 #include <linux/module.h>
18 #include <linux/kernel.h>
19 #include <linux/init.h>
20 #include <linux/fs.h>
21 #include <linux/netdevice.h>
22 #include <linux/etherdevice.h>
23 #include <linux/interrupt.h>
24 #include <linux/ip.h>
25 #include <linux/ipv6.h>
26 #include <linux/mm.h>
27 #include <linux/overflow.h>
28 #include <linux/page_ref.h>
29 #include <linux/pci.h>
30 #include <linux/pci_regs.h>
31 #include <linux/msi.h>
32 #include <linux/ethtool.h>
33 #include <linux/log2.h>
34 #include <linux/if_vlan.h>
35 #include <linux/random.h>
36 #include <linux/vmalloc.h>
37 #include <linux/ktime.h>
38 
39 #include <net/tls.h>
40 #include <net/vxlan.h>
41 
42 #include "nfpcore/nfp_nsp.h"
43 #include "ccm.h"
44 #include "nfp_app.h"
45 #include "nfp_net_ctrl.h"
46 #include "nfp_net.h"
47 #include "nfp_net_sriov.h"
48 #include "nfp_port.h"
49 #include "crypto/crypto.h"
50 #include "crypto/fw.h"
51 
52 /**
53  * nfp_net_get_fw_version() - Read and parse the FW version
54  * @fw_ver:	Output fw_version structure to read to
55  * @ctrl_bar:	Mapped address of the control BAR
56  */
57 void nfp_net_get_fw_version(struct nfp_net_fw_version *fw_ver,
58 			    void __iomem *ctrl_bar)
59 {
60 	u32 reg;
61 
62 	reg = readl(ctrl_bar + NFP_NET_CFG_VERSION);
63 	put_unaligned_le32(reg, fw_ver);
64 }
65 
66 static dma_addr_t nfp_net_dma_map_rx(struct nfp_net_dp *dp, void *frag)
67 {
68 	return dma_map_single_attrs(dp->dev, frag + NFP_NET_RX_BUF_HEADROOM,
69 				    dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
70 				    dp->rx_dma_dir, DMA_ATTR_SKIP_CPU_SYNC);
71 }
72 
73 static void
74 nfp_net_dma_sync_dev_rx(const struct nfp_net_dp *dp, dma_addr_t dma_addr)
75 {
76 	dma_sync_single_for_device(dp->dev, dma_addr,
77 				   dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
78 				   dp->rx_dma_dir);
79 }
80 
81 static void nfp_net_dma_unmap_rx(struct nfp_net_dp *dp, dma_addr_t dma_addr)
82 {
83 	dma_unmap_single_attrs(dp->dev, dma_addr,
84 			       dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA,
85 			       dp->rx_dma_dir, DMA_ATTR_SKIP_CPU_SYNC);
86 }
87 
88 static void nfp_net_dma_sync_cpu_rx(struct nfp_net_dp *dp, dma_addr_t dma_addr,
89 				    unsigned int len)
90 {
91 	dma_sync_single_for_cpu(dp->dev, dma_addr - NFP_NET_RX_BUF_HEADROOM,
92 				len, dp->rx_dma_dir);
93 }
94 
95 /* Firmware reconfig
96  *
97  * Firmware reconfig may take a while so we have two versions of it -
98  * synchronous and asynchronous (posted).  All synchronous callers are holding
99  * RTNL so we don't have to worry about serializing them.
100  */
101 static void nfp_net_reconfig_start(struct nfp_net *nn, u32 update)
102 {
103 	nn_writel(nn, NFP_NET_CFG_UPDATE, update);
104 	/* ensure update is written before pinging HW */
105 	nn_pci_flush(nn);
106 	nfp_qcp_wr_ptr_add(nn->qcp_cfg, 1);
107 	nn->reconfig_in_progress_update = update;
108 }
109 
110 /* Pass 0 as update to run posted reconfigs. */
111 static void nfp_net_reconfig_start_async(struct nfp_net *nn, u32 update)
112 {
113 	update |= nn->reconfig_posted;
114 	nn->reconfig_posted = 0;
115 
116 	nfp_net_reconfig_start(nn, update);
117 
118 	nn->reconfig_timer_active = true;
119 	mod_timer(&nn->reconfig_timer, jiffies + NFP_NET_POLL_TIMEOUT * HZ);
120 }
121 
122 static bool nfp_net_reconfig_check_done(struct nfp_net *nn, bool last_check)
123 {
124 	u32 reg;
125 
126 	reg = nn_readl(nn, NFP_NET_CFG_UPDATE);
127 	if (reg == 0)
128 		return true;
129 	if (reg & NFP_NET_CFG_UPDATE_ERR) {
130 		nn_err(nn, "Reconfig error (status: 0x%08x update: 0x%08x ctrl: 0x%08x)\n",
131 		       reg, nn->reconfig_in_progress_update,
132 		       nn_readl(nn, NFP_NET_CFG_CTRL));
133 		return true;
134 	} else if (last_check) {
135 		nn_err(nn, "Reconfig timeout (status: 0x%08x update: 0x%08x ctrl: 0x%08x)\n",
136 		       reg, nn->reconfig_in_progress_update,
137 		       nn_readl(nn, NFP_NET_CFG_CTRL));
138 		return true;
139 	}
140 
141 	return false;
142 }
143 
144 static bool __nfp_net_reconfig_wait(struct nfp_net *nn, unsigned long deadline)
145 {
146 	bool timed_out = false;
147 	int i;
148 
149 	/* Poll update field, waiting for NFP to ack the config.
150 	 * Do an opportunistic wait-busy loop, afterward sleep.
151 	 */
152 	for (i = 0; i < 50; i++) {
153 		if (nfp_net_reconfig_check_done(nn, false))
154 			return false;
155 		udelay(4);
156 	}
157 
158 	while (!nfp_net_reconfig_check_done(nn, timed_out)) {
159 		usleep_range(250, 500);
160 		timed_out = time_is_before_eq_jiffies(deadline);
161 	}
162 
163 	return timed_out;
164 }
165 
166 static int nfp_net_reconfig_wait(struct nfp_net *nn, unsigned long deadline)
167 {
168 	if (__nfp_net_reconfig_wait(nn, deadline))
169 		return -EIO;
170 
171 	if (nn_readl(nn, NFP_NET_CFG_UPDATE) & NFP_NET_CFG_UPDATE_ERR)
172 		return -EIO;
173 
174 	return 0;
175 }
176 
177 static void nfp_net_reconfig_timer(struct timer_list *t)
178 {
179 	struct nfp_net *nn = from_timer(nn, t, reconfig_timer);
180 
181 	spin_lock_bh(&nn->reconfig_lock);
182 
183 	nn->reconfig_timer_active = false;
184 
185 	/* If sync caller is present it will take over from us */
186 	if (nn->reconfig_sync_present)
187 		goto done;
188 
189 	/* Read reconfig status and report errors */
190 	nfp_net_reconfig_check_done(nn, true);
191 
192 	if (nn->reconfig_posted)
193 		nfp_net_reconfig_start_async(nn, 0);
194 done:
195 	spin_unlock_bh(&nn->reconfig_lock);
196 }
197 
198 /**
199  * nfp_net_reconfig_post() - Post async reconfig request
200  * @nn:      NFP Net device to reconfigure
201  * @update:  The value for the update field in the BAR config
202  *
203  * Record FW reconfiguration request.  Reconfiguration will be kicked off
204  * whenever reconfiguration machinery is idle.  Multiple requests can be
205  * merged together!
206  */
207 static void nfp_net_reconfig_post(struct nfp_net *nn, u32 update)
208 {
209 	spin_lock_bh(&nn->reconfig_lock);
210 
211 	/* Sync caller will kick off async reconf when it's done, just post */
212 	if (nn->reconfig_sync_present) {
213 		nn->reconfig_posted |= update;
214 		goto done;
215 	}
216 
217 	/* Opportunistically check if the previous command is done */
218 	if (!nn->reconfig_timer_active ||
219 	    nfp_net_reconfig_check_done(nn, false))
220 		nfp_net_reconfig_start_async(nn, update);
221 	else
222 		nn->reconfig_posted |= update;
223 done:
224 	spin_unlock_bh(&nn->reconfig_lock);
225 }
226 
227 static void nfp_net_reconfig_sync_enter(struct nfp_net *nn)
228 {
229 	bool cancelled_timer = false;
230 	u32 pre_posted_requests;
231 
232 	spin_lock_bh(&nn->reconfig_lock);
233 
234 	WARN_ON(nn->reconfig_sync_present);
235 	nn->reconfig_sync_present = true;
236 
237 	if (nn->reconfig_timer_active) {
238 		nn->reconfig_timer_active = false;
239 		cancelled_timer = true;
240 	}
241 	pre_posted_requests = nn->reconfig_posted;
242 	nn->reconfig_posted = 0;
243 
244 	spin_unlock_bh(&nn->reconfig_lock);
245 
246 	if (cancelled_timer) {
247 		del_timer_sync(&nn->reconfig_timer);
248 		nfp_net_reconfig_wait(nn, nn->reconfig_timer.expires);
249 	}
250 
251 	/* Run the posted reconfigs which were issued before we started */
252 	if (pre_posted_requests) {
253 		nfp_net_reconfig_start(nn, pre_posted_requests);
254 		nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
255 	}
256 }
257 
258 static void nfp_net_reconfig_wait_posted(struct nfp_net *nn)
259 {
260 	nfp_net_reconfig_sync_enter(nn);
261 
262 	spin_lock_bh(&nn->reconfig_lock);
263 	nn->reconfig_sync_present = false;
264 	spin_unlock_bh(&nn->reconfig_lock);
265 }
266 
267 /**
268  * __nfp_net_reconfig() - Reconfigure the firmware
269  * @nn:      NFP Net device to reconfigure
270  * @update:  The value for the update field in the BAR config
271  *
272  * Write the update word to the BAR and ping the reconfig queue.  The
273  * poll until the firmware has acknowledged the update by zeroing the
274  * update word.
275  *
276  * Return: Negative errno on error, 0 on success
277  */
278 int __nfp_net_reconfig(struct nfp_net *nn, u32 update)
279 {
280 	int ret;
281 
282 	nfp_net_reconfig_sync_enter(nn);
283 
284 	nfp_net_reconfig_start(nn, update);
285 	ret = nfp_net_reconfig_wait(nn, jiffies + HZ * NFP_NET_POLL_TIMEOUT);
286 
287 	spin_lock_bh(&nn->reconfig_lock);
288 
289 	if (nn->reconfig_posted)
290 		nfp_net_reconfig_start_async(nn, 0);
291 
292 	nn->reconfig_sync_present = false;
293 
294 	spin_unlock_bh(&nn->reconfig_lock);
295 
296 	return ret;
297 }
298 
299 int nfp_net_reconfig(struct nfp_net *nn, u32 update)
300 {
301 	int ret;
302 
303 	nn_ctrl_bar_lock(nn);
304 	ret = __nfp_net_reconfig(nn, update);
305 	nn_ctrl_bar_unlock(nn);
306 
307 	return ret;
308 }
309 
310 int nfp_net_mbox_lock(struct nfp_net *nn, unsigned int data_size)
311 {
312 	if (nn->tlv_caps.mbox_len < NFP_NET_CFG_MBOX_SIMPLE_VAL + data_size) {
313 		nn_err(nn, "mailbox too small for %u of data (%u)\n",
314 		       data_size, nn->tlv_caps.mbox_len);
315 		return -EIO;
316 	}
317 
318 	nn_ctrl_bar_lock(nn);
319 	return 0;
320 }
321 
322 /**
323  * nfp_net_mbox_reconfig() - Reconfigure the firmware via the mailbox
324  * @nn:        NFP Net device to reconfigure
325  * @mbox_cmd:  The value for the mailbox command
326  *
327  * Helper function for mailbox updates
328  *
329  * Return: Negative errno on error, 0 on success
330  */
331 int nfp_net_mbox_reconfig(struct nfp_net *nn, u32 mbox_cmd)
332 {
333 	u32 mbox = nn->tlv_caps.mbox_off;
334 	int ret;
335 
336 	nn_writeq(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_CMD, mbox_cmd);
337 
338 	ret = __nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MBOX);
339 	if (ret) {
340 		nn_err(nn, "Mailbox update error\n");
341 		return ret;
342 	}
343 
344 	return -nn_readl(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_RET);
345 }
346 
347 void nfp_net_mbox_reconfig_post(struct nfp_net *nn, u32 mbox_cmd)
348 {
349 	u32 mbox = nn->tlv_caps.mbox_off;
350 
351 	nn_writeq(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_CMD, mbox_cmd);
352 
353 	nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_MBOX);
354 }
355 
356 int nfp_net_mbox_reconfig_wait_posted(struct nfp_net *nn)
357 {
358 	u32 mbox = nn->tlv_caps.mbox_off;
359 
360 	nfp_net_reconfig_wait_posted(nn);
361 
362 	return -nn_readl(nn, mbox + NFP_NET_CFG_MBOX_SIMPLE_RET);
363 }
364 
365 int nfp_net_mbox_reconfig_and_unlock(struct nfp_net *nn, u32 mbox_cmd)
366 {
367 	int ret;
368 
369 	ret = nfp_net_mbox_reconfig(nn, mbox_cmd);
370 	nn_ctrl_bar_unlock(nn);
371 	return ret;
372 }
373 
374 /* Interrupt configuration and handling
375  */
376 
377 /**
378  * nfp_net_irq_unmask() - Unmask automasked interrupt
379  * @nn:       NFP Network structure
380  * @entry_nr: MSI-X table entry
381  *
382  * Clear the ICR for the IRQ entry.
383  */
384 static void nfp_net_irq_unmask(struct nfp_net *nn, unsigned int entry_nr)
385 {
386 	nn_writeb(nn, NFP_NET_CFG_ICR(entry_nr), NFP_NET_CFG_ICR_UNMASKED);
387 	nn_pci_flush(nn);
388 }
389 
390 /**
391  * nfp_net_irqs_alloc() - allocates MSI-X irqs
392  * @pdev:        PCI device structure
393  * @irq_entries: Array to be initialized and used to hold the irq entries
394  * @min_irqs:    Minimal acceptable number of interrupts
395  * @wanted_irqs: Target number of interrupts to allocate
396  *
397  * Return: Number of irqs obtained or 0 on error.
398  */
399 unsigned int
400 nfp_net_irqs_alloc(struct pci_dev *pdev, struct msix_entry *irq_entries,
401 		   unsigned int min_irqs, unsigned int wanted_irqs)
402 {
403 	unsigned int i;
404 	int got_irqs;
405 
406 	for (i = 0; i < wanted_irqs; i++)
407 		irq_entries[i].entry = i;
408 
409 	got_irqs = pci_enable_msix_range(pdev, irq_entries,
410 					 min_irqs, wanted_irqs);
411 	if (got_irqs < 0) {
412 		dev_err(&pdev->dev, "Failed to enable %d-%d MSI-X (err=%d)\n",
413 			min_irqs, wanted_irqs, got_irqs);
414 		return 0;
415 	}
416 
417 	if (got_irqs < wanted_irqs)
418 		dev_warn(&pdev->dev, "Unable to allocate %d IRQs got only %d\n",
419 			 wanted_irqs, got_irqs);
420 
421 	return got_irqs;
422 }
423 
424 /**
425  * nfp_net_irqs_assign() - Assign interrupts allocated externally to netdev
426  * @nn:		 NFP Network structure
427  * @irq_entries: Table of allocated interrupts
428  * @n:		 Size of @irq_entries (number of entries to grab)
429  *
430  * After interrupts are allocated with nfp_net_irqs_alloc() this function
431  * should be called to assign them to a specific netdev (port).
432  */
433 void
434 nfp_net_irqs_assign(struct nfp_net *nn, struct msix_entry *irq_entries,
435 		    unsigned int n)
436 {
437 	struct nfp_net_dp *dp = &nn->dp;
438 
439 	nn->max_r_vecs = n - NFP_NET_NON_Q_VECTORS;
440 	dp->num_r_vecs = nn->max_r_vecs;
441 
442 	memcpy(nn->irq_entries, irq_entries, sizeof(*irq_entries) * n);
443 
444 	if (dp->num_rx_rings > dp->num_r_vecs ||
445 	    dp->num_tx_rings > dp->num_r_vecs)
446 		dev_warn(nn->dp.dev, "More rings (%d,%d) than vectors (%d).\n",
447 			 dp->num_rx_rings, dp->num_tx_rings,
448 			 dp->num_r_vecs);
449 
450 	dp->num_rx_rings = min(dp->num_r_vecs, dp->num_rx_rings);
451 	dp->num_tx_rings = min(dp->num_r_vecs, dp->num_tx_rings);
452 	dp->num_stack_tx_rings = dp->num_tx_rings;
453 }
454 
455 /**
456  * nfp_net_irqs_disable() - Disable interrupts
457  * @pdev:        PCI device structure
458  *
459  * Undoes what @nfp_net_irqs_alloc() does.
460  */
461 void nfp_net_irqs_disable(struct pci_dev *pdev)
462 {
463 	pci_disable_msix(pdev);
464 }
465 
466 /**
467  * nfp_net_irq_rxtx() - Interrupt service routine for RX/TX rings.
468  * @irq:      Interrupt
469  * @data:     Opaque data structure
470  *
471  * Return: Indicate if the interrupt has been handled.
472  */
473 static irqreturn_t nfp_net_irq_rxtx(int irq, void *data)
474 {
475 	struct nfp_net_r_vector *r_vec = data;
476 
477 	napi_schedule_irqoff(&r_vec->napi);
478 
479 	/* The FW auto-masks any interrupt, either via the MASK bit in
480 	 * the MSI-X table or via the per entry ICR field.  So there
481 	 * is no need to disable interrupts here.
482 	 */
483 	return IRQ_HANDLED;
484 }
485 
486 static irqreturn_t nfp_ctrl_irq_rxtx(int irq, void *data)
487 {
488 	struct nfp_net_r_vector *r_vec = data;
489 
490 	tasklet_schedule(&r_vec->tasklet);
491 
492 	return IRQ_HANDLED;
493 }
494 
495 /**
496  * nfp_net_read_link_status() - Reread link status from control BAR
497  * @nn:       NFP Network structure
498  */
499 static void nfp_net_read_link_status(struct nfp_net *nn)
500 {
501 	unsigned long flags;
502 	bool link_up;
503 	u32 sts;
504 
505 	spin_lock_irqsave(&nn->link_status_lock, flags);
506 
507 	sts = nn_readl(nn, NFP_NET_CFG_STS);
508 	link_up = !!(sts & NFP_NET_CFG_STS_LINK);
509 
510 	if (nn->link_up == link_up)
511 		goto out;
512 
513 	nn->link_up = link_up;
514 	if (nn->port)
515 		set_bit(NFP_PORT_CHANGED, &nn->port->flags);
516 
517 	if (nn->link_up) {
518 		netif_carrier_on(nn->dp.netdev);
519 		netdev_info(nn->dp.netdev, "NIC Link is Up\n");
520 	} else {
521 		netif_carrier_off(nn->dp.netdev);
522 		netdev_info(nn->dp.netdev, "NIC Link is Down\n");
523 	}
524 out:
525 	spin_unlock_irqrestore(&nn->link_status_lock, flags);
526 }
527 
528 /**
529  * nfp_net_irq_lsc() - Interrupt service routine for link state changes
530  * @irq:      Interrupt
531  * @data:     Opaque data structure
532  *
533  * Return: Indicate if the interrupt has been handled.
534  */
535 static irqreturn_t nfp_net_irq_lsc(int irq, void *data)
536 {
537 	struct nfp_net *nn = data;
538 	struct msix_entry *entry;
539 
540 	entry = &nn->irq_entries[NFP_NET_IRQ_LSC_IDX];
541 
542 	nfp_net_read_link_status(nn);
543 
544 	nfp_net_irq_unmask(nn, entry->entry);
545 
546 	return IRQ_HANDLED;
547 }
548 
549 /**
550  * nfp_net_irq_exn() - Interrupt service routine for exceptions
551  * @irq:      Interrupt
552  * @data:     Opaque data structure
553  *
554  * Return: Indicate if the interrupt has been handled.
555  */
556 static irqreturn_t nfp_net_irq_exn(int irq, void *data)
557 {
558 	struct nfp_net *nn = data;
559 
560 	nn_err(nn, "%s: UNIMPLEMENTED.\n", __func__);
561 	/* XXX TO BE IMPLEMENTED */
562 	return IRQ_HANDLED;
563 }
564 
565 /**
566  * nfp_net_tx_ring_init() - Fill in the boilerplate for a TX ring
567  * @tx_ring:  TX ring structure
568  * @r_vec:    IRQ vector servicing this ring
569  * @idx:      Ring index
570  * @is_xdp:   Is this an XDP TX ring?
571  */
572 static void
573 nfp_net_tx_ring_init(struct nfp_net_tx_ring *tx_ring,
574 		     struct nfp_net_r_vector *r_vec, unsigned int idx,
575 		     bool is_xdp)
576 {
577 	struct nfp_net *nn = r_vec->nfp_net;
578 
579 	tx_ring->idx = idx;
580 	tx_ring->r_vec = r_vec;
581 	tx_ring->is_xdp = is_xdp;
582 	u64_stats_init(&tx_ring->r_vec->tx_sync);
583 
584 	tx_ring->qcidx = tx_ring->idx * nn->stride_tx;
585 	tx_ring->qcp_q = nn->tx_bar + NFP_QCP_QUEUE_OFF(tx_ring->qcidx);
586 }
587 
588 /**
589  * nfp_net_rx_ring_init() - Fill in the boilerplate for a RX ring
590  * @rx_ring:  RX ring structure
591  * @r_vec:    IRQ vector servicing this ring
592  * @idx:      Ring index
593  */
594 static void
595 nfp_net_rx_ring_init(struct nfp_net_rx_ring *rx_ring,
596 		     struct nfp_net_r_vector *r_vec, unsigned int idx)
597 {
598 	struct nfp_net *nn = r_vec->nfp_net;
599 
600 	rx_ring->idx = idx;
601 	rx_ring->r_vec = r_vec;
602 	u64_stats_init(&rx_ring->r_vec->rx_sync);
603 
604 	rx_ring->fl_qcidx = rx_ring->idx * nn->stride_rx;
605 	rx_ring->qcp_fl = nn->rx_bar + NFP_QCP_QUEUE_OFF(rx_ring->fl_qcidx);
606 }
607 
608 /**
609  * nfp_net_aux_irq_request() - Request an auxiliary interrupt (LSC or EXN)
610  * @nn:		NFP Network structure
611  * @ctrl_offset: Control BAR offset where IRQ configuration should be written
612  * @format:	printf-style format to construct the interrupt name
613  * @name:	Pointer to allocated space for interrupt name
614  * @name_sz:	Size of space for interrupt name
615  * @vector_idx:	Index of MSI-X vector used for this interrupt
616  * @handler:	IRQ handler to register for this interrupt
617  */
618 static int
619 nfp_net_aux_irq_request(struct nfp_net *nn, u32 ctrl_offset,
620 			const char *format, char *name, size_t name_sz,
621 			unsigned int vector_idx, irq_handler_t handler)
622 {
623 	struct msix_entry *entry;
624 	int err;
625 
626 	entry = &nn->irq_entries[vector_idx];
627 
628 	snprintf(name, name_sz, format, nfp_net_name(nn));
629 	err = request_irq(entry->vector, handler, 0, name, nn);
630 	if (err) {
631 		nn_err(nn, "Failed to request IRQ %d (err=%d).\n",
632 		       entry->vector, err);
633 		return err;
634 	}
635 	nn_writeb(nn, ctrl_offset, entry->entry);
636 	nfp_net_irq_unmask(nn, entry->entry);
637 
638 	return 0;
639 }
640 
641 /**
642  * nfp_net_aux_irq_free() - Free an auxiliary interrupt (LSC or EXN)
643  * @nn:		NFP Network structure
644  * @ctrl_offset: Control BAR offset where IRQ configuration should be written
645  * @vector_idx:	Index of MSI-X vector used for this interrupt
646  */
647 static void nfp_net_aux_irq_free(struct nfp_net *nn, u32 ctrl_offset,
648 				 unsigned int vector_idx)
649 {
650 	nn_writeb(nn, ctrl_offset, 0xff);
651 	nn_pci_flush(nn);
652 	free_irq(nn->irq_entries[vector_idx].vector, nn);
653 }
654 
655 /* Transmit
656  *
657  * One queue controller peripheral queue is used for transmit.  The
658  * driver en-queues packets for transmit by advancing the write
659  * pointer.  The device indicates that packets have transmitted by
660  * advancing the read pointer.  The driver maintains a local copy of
661  * the read and write pointer in @struct nfp_net_tx_ring.  The driver
662  * keeps @wr_p in sync with the queue controller write pointer and can
663  * determine how many packets have been transmitted by comparing its
664  * copy of the read pointer @rd_p with the read pointer maintained by
665  * the queue controller peripheral.
666  */
667 
668 /**
669  * nfp_net_tx_full() - Check if the TX ring is full
670  * @tx_ring: TX ring to check
671  * @dcnt:    Number of descriptors that need to be enqueued (must be >= 1)
672  *
673  * This function checks, based on the *host copy* of read/write
674  * pointer if a given TX ring is full.  The real TX queue may have
675  * some newly made available slots.
676  *
677  * Return: True if the ring is full.
678  */
679 static int nfp_net_tx_full(struct nfp_net_tx_ring *tx_ring, int dcnt)
680 {
681 	return (tx_ring->wr_p - tx_ring->rd_p) >= (tx_ring->cnt - dcnt);
682 }
683 
684 /* Wrappers for deciding when to stop and restart TX queues */
685 static int nfp_net_tx_ring_should_wake(struct nfp_net_tx_ring *tx_ring)
686 {
687 	return !nfp_net_tx_full(tx_ring, MAX_SKB_FRAGS * 4);
688 }
689 
690 static int nfp_net_tx_ring_should_stop(struct nfp_net_tx_ring *tx_ring)
691 {
692 	return nfp_net_tx_full(tx_ring, MAX_SKB_FRAGS + 1);
693 }
694 
695 /**
696  * nfp_net_tx_ring_stop() - stop tx ring
697  * @nd_q:    netdev queue
698  * @tx_ring: driver tx queue structure
699  *
700  * Safely stop TX ring.  Remember that while we are running .start_xmit()
701  * someone else may be cleaning the TX ring completions so we need to be
702  * extra careful here.
703  */
704 static void nfp_net_tx_ring_stop(struct netdev_queue *nd_q,
705 				 struct nfp_net_tx_ring *tx_ring)
706 {
707 	netif_tx_stop_queue(nd_q);
708 
709 	/* We can race with the TX completion out of NAPI so recheck */
710 	smp_mb();
711 	if (unlikely(nfp_net_tx_ring_should_wake(tx_ring)))
712 		netif_tx_start_queue(nd_q);
713 }
714 
715 /**
716  * nfp_net_tx_tso() - Set up Tx descriptor for LSO
717  * @r_vec: per-ring structure
718  * @txbuf: Pointer to driver soft TX descriptor
719  * @txd: Pointer to HW TX descriptor
720  * @skb: Pointer to SKB
721  * @md_bytes: Prepend length
722  *
723  * Set up Tx descriptor for LSO, do nothing for non-LSO skbs.
724  * Return error on packet header greater than maximum supported LSO header size.
725  */
726 static void nfp_net_tx_tso(struct nfp_net_r_vector *r_vec,
727 			   struct nfp_net_tx_buf *txbuf,
728 			   struct nfp_net_tx_desc *txd, struct sk_buff *skb,
729 			   u32 md_bytes)
730 {
731 	u32 l3_offset, l4_offset, hdrlen;
732 	u16 mss;
733 
734 	if (!skb_is_gso(skb))
735 		return;
736 
737 	if (!skb->encapsulation) {
738 		l3_offset = skb_network_offset(skb);
739 		l4_offset = skb_transport_offset(skb);
740 		hdrlen = skb_transport_offset(skb) + tcp_hdrlen(skb);
741 	} else {
742 		l3_offset = skb_inner_network_offset(skb);
743 		l4_offset = skb_inner_transport_offset(skb);
744 		hdrlen = skb_inner_transport_header(skb) - skb->data +
745 			inner_tcp_hdrlen(skb);
746 	}
747 
748 	txbuf->pkt_cnt = skb_shinfo(skb)->gso_segs;
749 	txbuf->real_len += hdrlen * (txbuf->pkt_cnt - 1);
750 
751 	mss = skb_shinfo(skb)->gso_size & PCIE_DESC_TX_MSS_MASK;
752 	txd->l3_offset = l3_offset - md_bytes;
753 	txd->l4_offset = l4_offset - md_bytes;
754 	txd->lso_hdrlen = hdrlen - md_bytes;
755 	txd->mss = cpu_to_le16(mss);
756 	txd->flags |= PCIE_DESC_TX_LSO;
757 
758 	u64_stats_update_begin(&r_vec->tx_sync);
759 	r_vec->tx_lso++;
760 	u64_stats_update_end(&r_vec->tx_sync);
761 }
762 
763 /**
764  * nfp_net_tx_csum() - Set TX CSUM offload flags in TX descriptor
765  * @dp:  NFP Net data path struct
766  * @r_vec: per-ring structure
767  * @txbuf: Pointer to driver soft TX descriptor
768  * @txd: Pointer to TX descriptor
769  * @skb: Pointer to SKB
770  *
771  * This function sets the TX checksum flags in the TX descriptor based
772  * on the configuration and the protocol of the packet to be transmitted.
773  */
774 static void nfp_net_tx_csum(struct nfp_net_dp *dp,
775 			    struct nfp_net_r_vector *r_vec,
776 			    struct nfp_net_tx_buf *txbuf,
777 			    struct nfp_net_tx_desc *txd, struct sk_buff *skb)
778 {
779 	struct ipv6hdr *ipv6h;
780 	struct iphdr *iph;
781 	u8 l4_hdr;
782 
783 	if (!(dp->ctrl & NFP_NET_CFG_CTRL_TXCSUM))
784 		return;
785 
786 	if (skb->ip_summed != CHECKSUM_PARTIAL)
787 		return;
788 
789 	txd->flags |= PCIE_DESC_TX_CSUM;
790 	if (skb->encapsulation)
791 		txd->flags |= PCIE_DESC_TX_ENCAP;
792 
793 	iph = skb->encapsulation ? inner_ip_hdr(skb) : ip_hdr(skb);
794 	ipv6h = skb->encapsulation ? inner_ipv6_hdr(skb) : ipv6_hdr(skb);
795 
796 	if (iph->version == 4) {
797 		txd->flags |= PCIE_DESC_TX_IP4_CSUM;
798 		l4_hdr = iph->protocol;
799 	} else if (ipv6h->version == 6) {
800 		l4_hdr = ipv6h->nexthdr;
801 	} else {
802 		nn_dp_warn(dp, "partial checksum but ipv=%x!\n", iph->version);
803 		return;
804 	}
805 
806 	switch (l4_hdr) {
807 	case IPPROTO_TCP:
808 		txd->flags |= PCIE_DESC_TX_TCP_CSUM;
809 		break;
810 	case IPPROTO_UDP:
811 		txd->flags |= PCIE_DESC_TX_UDP_CSUM;
812 		break;
813 	default:
814 		nn_dp_warn(dp, "partial checksum but l4 proto=%x!\n", l4_hdr);
815 		return;
816 	}
817 
818 	u64_stats_update_begin(&r_vec->tx_sync);
819 	if (skb->encapsulation)
820 		r_vec->hw_csum_tx_inner += txbuf->pkt_cnt;
821 	else
822 		r_vec->hw_csum_tx += txbuf->pkt_cnt;
823 	u64_stats_update_end(&r_vec->tx_sync);
824 }
825 
826 static struct sk_buff *
827 nfp_net_tls_tx(struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec,
828 	       struct sk_buff *skb, u64 *tls_handle, int *nr_frags)
829 {
830 #ifdef CONFIG_TLS_DEVICE
831 	struct nfp_net_tls_offload_ctx *ntls;
832 	struct sk_buff *nskb;
833 	bool resync_pending;
834 	u32 datalen, seq;
835 
836 	if (likely(!dp->ktls_tx))
837 		return skb;
838 	if (!skb->sk || !tls_is_sk_tx_device_offloaded(skb->sk))
839 		return skb;
840 
841 	datalen = skb->len - (skb_transport_offset(skb) + tcp_hdrlen(skb));
842 	seq = ntohl(tcp_hdr(skb)->seq);
843 	ntls = tls_driver_ctx(skb->sk, TLS_OFFLOAD_CTX_DIR_TX);
844 	resync_pending = tls_offload_tx_resync_pending(skb->sk);
845 	if (unlikely(resync_pending || ntls->next_seq != seq)) {
846 		/* Pure ACK out of order already */
847 		if (!datalen)
848 			return skb;
849 
850 		u64_stats_update_begin(&r_vec->tx_sync);
851 		r_vec->tls_tx_fallback++;
852 		u64_stats_update_end(&r_vec->tx_sync);
853 
854 		nskb = tls_encrypt_skb(skb);
855 		if (!nskb) {
856 			u64_stats_update_begin(&r_vec->tx_sync);
857 			r_vec->tls_tx_no_fallback++;
858 			u64_stats_update_end(&r_vec->tx_sync);
859 			return NULL;
860 		}
861 		/* encryption wasn't necessary */
862 		if (nskb == skb)
863 			return skb;
864 		/* we don't re-check ring space */
865 		if (unlikely(skb_is_nonlinear(nskb))) {
866 			nn_dp_warn(dp, "tls_encrypt_skb() produced fragmented frame\n");
867 			u64_stats_update_begin(&r_vec->tx_sync);
868 			r_vec->tx_errors++;
869 			u64_stats_update_end(&r_vec->tx_sync);
870 			dev_kfree_skb_any(nskb);
871 			return NULL;
872 		}
873 
874 		/* jump forward, a TX may have gotten lost, need to sync TX */
875 		if (!resync_pending && seq - ntls->next_seq < U32_MAX / 4)
876 			tls_offload_tx_resync_request(nskb->sk, seq,
877 						      ntls->next_seq);
878 
879 		*nr_frags = 0;
880 		return nskb;
881 	}
882 
883 	if (datalen) {
884 		u64_stats_update_begin(&r_vec->tx_sync);
885 		if (!skb_is_gso(skb))
886 			r_vec->hw_tls_tx++;
887 		else
888 			r_vec->hw_tls_tx += skb_shinfo(skb)->gso_segs;
889 		u64_stats_update_end(&r_vec->tx_sync);
890 	}
891 
892 	memcpy(tls_handle, ntls->fw_handle, sizeof(ntls->fw_handle));
893 	ntls->next_seq += datalen;
894 #endif
895 	return skb;
896 }
897 
898 static void nfp_net_tls_tx_undo(struct sk_buff *skb, u64 tls_handle)
899 {
900 #ifdef CONFIG_TLS_DEVICE
901 	struct nfp_net_tls_offload_ctx *ntls;
902 	u32 datalen, seq;
903 
904 	if (!tls_handle)
905 		return;
906 	if (WARN_ON_ONCE(!skb->sk || !tls_is_sk_tx_device_offloaded(skb->sk)))
907 		return;
908 
909 	datalen = skb->len - (skb_transport_offset(skb) + tcp_hdrlen(skb));
910 	seq = ntohl(tcp_hdr(skb)->seq);
911 
912 	ntls = tls_driver_ctx(skb->sk, TLS_OFFLOAD_CTX_DIR_TX);
913 	if (ntls->next_seq == seq + datalen)
914 		ntls->next_seq = seq;
915 	else
916 		WARN_ON_ONCE(1);
917 #endif
918 }
919 
920 static void nfp_net_tx_xmit_more_flush(struct nfp_net_tx_ring *tx_ring)
921 {
922 	wmb();
923 	nfp_qcp_wr_ptr_add(tx_ring->qcp_q, tx_ring->wr_ptr_add);
924 	tx_ring->wr_ptr_add = 0;
925 }
926 
927 static int nfp_net_prep_tx_meta(struct sk_buff *skb, u64 tls_handle)
928 {
929 	struct metadata_dst *md_dst = skb_metadata_dst(skb);
930 	unsigned char *data;
931 	u32 meta_id = 0;
932 	int md_bytes;
933 
934 	if (likely(!md_dst && !tls_handle))
935 		return 0;
936 	if (unlikely(md_dst && md_dst->type != METADATA_HW_PORT_MUX)) {
937 		if (!tls_handle)
938 			return 0;
939 		md_dst = NULL;
940 	}
941 
942 	md_bytes = 4 + !!md_dst * 4 + !!tls_handle * 8;
943 
944 	if (unlikely(skb_cow_head(skb, md_bytes)))
945 		return -ENOMEM;
946 
947 	meta_id = 0;
948 	data = skb_push(skb, md_bytes) + md_bytes;
949 	if (md_dst) {
950 		data -= 4;
951 		put_unaligned_be32(md_dst->u.port_info.port_id, data);
952 		meta_id = NFP_NET_META_PORTID;
953 	}
954 	if (tls_handle) {
955 		/* conn handle is opaque, we just use u64 to be able to quickly
956 		 * compare it to zero
957 		 */
958 		data -= 8;
959 		memcpy(data, &tls_handle, sizeof(tls_handle));
960 		meta_id <<= NFP_NET_META_FIELD_SIZE;
961 		meta_id |= NFP_NET_META_CONN_HANDLE;
962 	}
963 
964 	data -= 4;
965 	put_unaligned_be32(meta_id, data);
966 
967 	return md_bytes;
968 }
969 
970 /**
971  * nfp_net_tx() - Main transmit entry point
972  * @skb:    SKB to transmit
973  * @netdev: netdev structure
974  *
975  * Return: NETDEV_TX_OK on success.
976  */
977 static netdev_tx_t nfp_net_tx(struct sk_buff *skb, struct net_device *netdev)
978 {
979 	struct nfp_net *nn = netdev_priv(netdev);
980 	const skb_frag_t *frag;
981 	int f, nr_frags, wr_idx, md_bytes;
982 	struct nfp_net_tx_ring *tx_ring;
983 	struct nfp_net_r_vector *r_vec;
984 	struct nfp_net_tx_buf *txbuf;
985 	struct nfp_net_tx_desc *txd;
986 	struct netdev_queue *nd_q;
987 	struct nfp_net_dp *dp;
988 	dma_addr_t dma_addr;
989 	unsigned int fsize;
990 	u64 tls_handle = 0;
991 	u16 qidx;
992 
993 	dp = &nn->dp;
994 	qidx = skb_get_queue_mapping(skb);
995 	tx_ring = &dp->tx_rings[qidx];
996 	r_vec = tx_ring->r_vec;
997 
998 	nr_frags = skb_shinfo(skb)->nr_frags;
999 
1000 	if (unlikely(nfp_net_tx_full(tx_ring, nr_frags + 1))) {
1001 		nn_dp_warn(dp, "TX ring %d busy. wrp=%u rdp=%u\n",
1002 			   qidx, tx_ring->wr_p, tx_ring->rd_p);
1003 		nd_q = netdev_get_tx_queue(dp->netdev, qidx);
1004 		netif_tx_stop_queue(nd_q);
1005 		nfp_net_tx_xmit_more_flush(tx_ring);
1006 		u64_stats_update_begin(&r_vec->tx_sync);
1007 		r_vec->tx_busy++;
1008 		u64_stats_update_end(&r_vec->tx_sync);
1009 		return NETDEV_TX_BUSY;
1010 	}
1011 
1012 	skb = nfp_net_tls_tx(dp, r_vec, skb, &tls_handle, &nr_frags);
1013 	if (unlikely(!skb)) {
1014 		nfp_net_tx_xmit_more_flush(tx_ring);
1015 		return NETDEV_TX_OK;
1016 	}
1017 
1018 	md_bytes = nfp_net_prep_tx_meta(skb, tls_handle);
1019 	if (unlikely(md_bytes < 0))
1020 		goto err_flush;
1021 
1022 	/* Start with the head skbuf */
1023 	dma_addr = dma_map_single(dp->dev, skb->data, skb_headlen(skb),
1024 				  DMA_TO_DEVICE);
1025 	if (dma_mapping_error(dp->dev, dma_addr))
1026 		goto err_dma_err;
1027 
1028 	wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
1029 
1030 	/* Stash the soft descriptor of the head then initialize it */
1031 	txbuf = &tx_ring->txbufs[wr_idx];
1032 	txbuf->skb = skb;
1033 	txbuf->dma_addr = dma_addr;
1034 	txbuf->fidx = -1;
1035 	txbuf->pkt_cnt = 1;
1036 	txbuf->real_len = skb->len;
1037 
1038 	/* Build TX descriptor */
1039 	txd = &tx_ring->txds[wr_idx];
1040 	txd->offset_eop = (nr_frags ? 0 : PCIE_DESC_TX_EOP) | md_bytes;
1041 	txd->dma_len = cpu_to_le16(skb_headlen(skb));
1042 	nfp_desc_set_dma_addr(txd, dma_addr);
1043 	txd->data_len = cpu_to_le16(skb->len);
1044 
1045 	txd->flags = 0;
1046 	txd->mss = 0;
1047 	txd->lso_hdrlen = 0;
1048 
1049 	/* Do not reorder - tso may adjust pkt cnt, vlan may override fields */
1050 	nfp_net_tx_tso(r_vec, txbuf, txd, skb, md_bytes);
1051 	nfp_net_tx_csum(dp, r_vec, txbuf, txd, skb);
1052 	if (skb_vlan_tag_present(skb) && dp->ctrl & NFP_NET_CFG_CTRL_TXVLAN) {
1053 		txd->flags |= PCIE_DESC_TX_VLAN;
1054 		txd->vlan = cpu_to_le16(skb_vlan_tag_get(skb));
1055 	}
1056 
1057 	/* Gather DMA */
1058 	if (nr_frags > 0) {
1059 		__le64 second_half;
1060 
1061 		/* all descs must match except for in addr, length and eop */
1062 		second_half = txd->vals8[1];
1063 
1064 		for (f = 0; f < nr_frags; f++) {
1065 			frag = &skb_shinfo(skb)->frags[f];
1066 			fsize = skb_frag_size(frag);
1067 
1068 			dma_addr = skb_frag_dma_map(dp->dev, frag, 0,
1069 						    fsize, DMA_TO_DEVICE);
1070 			if (dma_mapping_error(dp->dev, dma_addr))
1071 				goto err_unmap;
1072 
1073 			wr_idx = D_IDX(tx_ring, wr_idx + 1);
1074 			tx_ring->txbufs[wr_idx].skb = skb;
1075 			tx_ring->txbufs[wr_idx].dma_addr = dma_addr;
1076 			tx_ring->txbufs[wr_idx].fidx = f;
1077 
1078 			txd = &tx_ring->txds[wr_idx];
1079 			txd->dma_len = cpu_to_le16(fsize);
1080 			nfp_desc_set_dma_addr(txd, dma_addr);
1081 			txd->offset_eop = md_bytes |
1082 				((f == nr_frags - 1) ? PCIE_DESC_TX_EOP : 0);
1083 			txd->vals8[1] = second_half;
1084 		}
1085 
1086 		u64_stats_update_begin(&r_vec->tx_sync);
1087 		r_vec->tx_gather++;
1088 		u64_stats_update_end(&r_vec->tx_sync);
1089 	}
1090 
1091 	skb_tx_timestamp(skb);
1092 
1093 	nd_q = netdev_get_tx_queue(dp->netdev, tx_ring->idx);
1094 
1095 	tx_ring->wr_p += nr_frags + 1;
1096 	if (nfp_net_tx_ring_should_stop(tx_ring))
1097 		nfp_net_tx_ring_stop(nd_q, tx_ring);
1098 
1099 	tx_ring->wr_ptr_add += nr_frags + 1;
1100 	if (__netdev_tx_sent_queue(nd_q, txbuf->real_len, netdev_xmit_more()))
1101 		nfp_net_tx_xmit_more_flush(tx_ring);
1102 
1103 	return NETDEV_TX_OK;
1104 
1105 err_unmap:
1106 	while (--f >= 0) {
1107 		frag = &skb_shinfo(skb)->frags[f];
1108 		dma_unmap_page(dp->dev, tx_ring->txbufs[wr_idx].dma_addr,
1109 			       skb_frag_size(frag), DMA_TO_DEVICE);
1110 		tx_ring->txbufs[wr_idx].skb = NULL;
1111 		tx_ring->txbufs[wr_idx].dma_addr = 0;
1112 		tx_ring->txbufs[wr_idx].fidx = -2;
1113 		wr_idx = wr_idx - 1;
1114 		if (wr_idx < 0)
1115 			wr_idx += tx_ring->cnt;
1116 	}
1117 	dma_unmap_single(dp->dev, tx_ring->txbufs[wr_idx].dma_addr,
1118 			 skb_headlen(skb), DMA_TO_DEVICE);
1119 	tx_ring->txbufs[wr_idx].skb = NULL;
1120 	tx_ring->txbufs[wr_idx].dma_addr = 0;
1121 	tx_ring->txbufs[wr_idx].fidx = -2;
1122 err_dma_err:
1123 	nn_dp_warn(dp, "Failed to map DMA TX buffer\n");
1124 err_flush:
1125 	nfp_net_tx_xmit_more_flush(tx_ring);
1126 	u64_stats_update_begin(&r_vec->tx_sync);
1127 	r_vec->tx_errors++;
1128 	u64_stats_update_end(&r_vec->tx_sync);
1129 	nfp_net_tls_tx_undo(skb, tls_handle);
1130 	dev_kfree_skb_any(skb);
1131 	return NETDEV_TX_OK;
1132 }
1133 
1134 /**
1135  * nfp_net_tx_complete() - Handled completed TX packets
1136  * @tx_ring:	TX ring structure
1137  * @budget:	NAPI budget (only used as bool to determine if in NAPI context)
1138  */
1139 static void nfp_net_tx_complete(struct nfp_net_tx_ring *tx_ring, int budget)
1140 {
1141 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
1142 	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
1143 	struct netdev_queue *nd_q;
1144 	u32 done_pkts = 0, done_bytes = 0;
1145 	u32 qcp_rd_p;
1146 	int todo;
1147 
1148 	if (tx_ring->wr_p == tx_ring->rd_p)
1149 		return;
1150 
1151 	/* Work out how many descriptors have been transmitted */
1152 	qcp_rd_p = nfp_qcp_rd_ptr_read(tx_ring->qcp_q);
1153 
1154 	if (qcp_rd_p == tx_ring->qcp_rd_p)
1155 		return;
1156 
1157 	todo = D_IDX(tx_ring, qcp_rd_p - tx_ring->qcp_rd_p);
1158 
1159 	while (todo--) {
1160 		const skb_frag_t *frag;
1161 		struct nfp_net_tx_buf *tx_buf;
1162 		struct sk_buff *skb;
1163 		int fidx, nr_frags;
1164 		int idx;
1165 
1166 		idx = D_IDX(tx_ring, tx_ring->rd_p++);
1167 		tx_buf = &tx_ring->txbufs[idx];
1168 
1169 		skb = tx_buf->skb;
1170 		if (!skb)
1171 			continue;
1172 
1173 		nr_frags = skb_shinfo(skb)->nr_frags;
1174 		fidx = tx_buf->fidx;
1175 
1176 		if (fidx == -1) {
1177 			/* unmap head */
1178 			dma_unmap_single(dp->dev, tx_buf->dma_addr,
1179 					 skb_headlen(skb), DMA_TO_DEVICE);
1180 
1181 			done_pkts += tx_buf->pkt_cnt;
1182 			done_bytes += tx_buf->real_len;
1183 		} else {
1184 			/* unmap fragment */
1185 			frag = &skb_shinfo(skb)->frags[fidx];
1186 			dma_unmap_page(dp->dev, tx_buf->dma_addr,
1187 				       skb_frag_size(frag), DMA_TO_DEVICE);
1188 		}
1189 
1190 		/* check for last gather fragment */
1191 		if (fidx == nr_frags - 1)
1192 			napi_consume_skb(skb, budget);
1193 
1194 		tx_buf->dma_addr = 0;
1195 		tx_buf->skb = NULL;
1196 		tx_buf->fidx = -2;
1197 	}
1198 
1199 	tx_ring->qcp_rd_p = qcp_rd_p;
1200 
1201 	u64_stats_update_begin(&r_vec->tx_sync);
1202 	r_vec->tx_bytes += done_bytes;
1203 	r_vec->tx_pkts += done_pkts;
1204 	u64_stats_update_end(&r_vec->tx_sync);
1205 
1206 	if (!dp->netdev)
1207 		return;
1208 
1209 	nd_q = netdev_get_tx_queue(dp->netdev, tx_ring->idx);
1210 	netdev_tx_completed_queue(nd_q, done_pkts, done_bytes);
1211 	if (nfp_net_tx_ring_should_wake(tx_ring)) {
1212 		/* Make sure TX thread will see updated tx_ring->rd_p */
1213 		smp_mb();
1214 
1215 		if (unlikely(netif_tx_queue_stopped(nd_q)))
1216 			netif_tx_wake_queue(nd_q);
1217 	}
1218 
1219 	WARN_ONCE(tx_ring->wr_p - tx_ring->rd_p > tx_ring->cnt,
1220 		  "TX ring corruption rd_p=%u wr_p=%u cnt=%u\n",
1221 		  tx_ring->rd_p, tx_ring->wr_p, tx_ring->cnt);
1222 }
1223 
1224 static bool nfp_net_xdp_complete(struct nfp_net_tx_ring *tx_ring)
1225 {
1226 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
1227 	u32 done_pkts = 0, done_bytes = 0;
1228 	bool done_all;
1229 	int idx, todo;
1230 	u32 qcp_rd_p;
1231 
1232 	/* Work out how many descriptors have been transmitted */
1233 	qcp_rd_p = nfp_qcp_rd_ptr_read(tx_ring->qcp_q);
1234 
1235 	if (qcp_rd_p == tx_ring->qcp_rd_p)
1236 		return true;
1237 
1238 	todo = D_IDX(tx_ring, qcp_rd_p - tx_ring->qcp_rd_p);
1239 
1240 	done_all = todo <= NFP_NET_XDP_MAX_COMPLETE;
1241 	todo = min(todo, NFP_NET_XDP_MAX_COMPLETE);
1242 
1243 	tx_ring->qcp_rd_p = D_IDX(tx_ring, tx_ring->qcp_rd_p + todo);
1244 
1245 	done_pkts = todo;
1246 	while (todo--) {
1247 		idx = D_IDX(tx_ring, tx_ring->rd_p);
1248 		tx_ring->rd_p++;
1249 
1250 		done_bytes += tx_ring->txbufs[idx].real_len;
1251 	}
1252 
1253 	u64_stats_update_begin(&r_vec->tx_sync);
1254 	r_vec->tx_bytes += done_bytes;
1255 	r_vec->tx_pkts += done_pkts;
1256 	u64_stats_update_end(&r_vec->tx_sync);
1257 
1258 	WARN_ONCE(tx_ring->wr_p - tx_ring->rd_p > tx_ring->cnt,
1259 		  "XDP TX ring corruption rd_p=%u wr_p=%u cnt=%u\n",
1260 		  tx_ring->rd_p, tx_ring->wr_p, tx_ring->cnt);
1261 
1262 	return done_all;
1263 }
1264 
1265 /**
1266  * nfp_net_tx_ring_reset() - Free any untransmitted buffers and reset pointers
1267  * @dp:		NFP Net data path struct
1268  * @tx_ring:	TX ring structure
1269  *
1270  * Assumes that the device is stopped, must be idempotent.
1271  */
1272 static void
1273 nfp_net_tx_ring_reset(struct nfp_net_dp *dp, struct nfp_net_tx_ring *tx_ring)
1274 {
1275 	const skb_frag_t *frag;
1276 	struct netdev_queue *nd_q;
1277 
1278 	while (!tx_ring->is_xdp && tx_ring->rd_p != tx_ring->wr_p) {
1279 		struct nfp_net_tx_buf *tx_buf;
1280 		struct sk_buff *skb;
1281 		int idx, nr_frags;
1282 
1283 		idx = D_IDX(tx_ring, tx_ring->rd_p);
1284 		tx_buf = &tx_ring->txbufs[idx];
1285 
1286 		skb = tx_ring->txbufs[idx].skb;
1287 		nr_frags = skb_shinfo(skb)->nr_frags;
1288 
1289 		if (tx_buf->fidx == -1) {
1290 			/* unmap head */
1291 			dma_unmap_single(dp->dev, tx_buf->dma_addr,
1292 					 skb_headlen(skb), DMA_TO_DEVICE);
1293 		} else {
1294 			/* unmap fragment */
1295 			frag = &skb_shinfo(skb)->frags[tx_buf->fidx];
1296 			dma_unmap_page(dp->dev, tx_buf->dma_addr,
1297 				       skb_frag_size(frag), DMA_TO_DEVICE);
1298 		}
1299 
1300 		/* check for last gather fragment */
1301 		if (tx_buf->fidx == nr_frags - 1)
1302 			dev_kfree_skb_any(skb);
1303 
1304 		tx_buf->dma_addr = 0;
1305 		tx_buf->skb = NULL;
1306 		tx_buf->fidx = -2;
1307 
1308 		tx_ring->qcp_rd_p++;
1309 		tx_ring->rd_p++;
1310 	}
1311 
1312 	memset(tx_ring->txds, 0, tx_ring->size);
1313 	tx_ring->wr_p = 0;
1314 	tx_ring->rd_p = 0;
1315 	tx_ring->qcp_rd_p = 0;
1316 	tx_ring->wr_ptr_add = 0;
1317 
1318 	if (tx_ring->is_xdp || !dp->netdev)
1319 		return;
1320 
1321 	nd_q = netdev_get_tx_queue(dp->netdev, tx_ring->idx);
1322 	netdev_tx_reset_queue(nd_q);
1323 }
1324 
1325 static void nfp_net_tx_timeout(struct net_device *netdev, unsigned int txqueue)
1326 {
1327 	struct nfp_net *nn = netdev_priv(netdev);
1328 
1329 	nn_warn(nn, "TX watchdog timeout on ring: %u\n", txqueue);
1330 }
1331 
1332 /* Receive processing
1333  */
1334 static unsigned int
1335 nfp_net_calc_fl_bufsz(struct nfp_net_dp *dp)
1336 {
1337 	unsigned int fl_bufsz;
1338 
1339 	fl_bufsz = NFP_NET_RX_BUF_HEADROOM;
1340 	fl_bufsz += dp->rx_dma_off;
1341 	if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
1342 		fl_bufsz += NFP_NET_MAX_PREPEND;
1343 	else
1344 		fl_bufsz += dp->rx_offset;
1345 	fl_bufsz += ETH_HLEN + VLAN_HLEN * 2 + dp->mtu;
1346 
1347 	fl_bufsz = SKB_DATA_ALIGN(fl_bufsz);
1348 	fl_bufsz += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
1349 
1350 	return fl_bufsz;
1351 }
1352 
1353 static void
1354 nfp_net_free_frag(void *frag, bool xdp)
1355 {
1356 	if (!xdp)
1357 		skb_free_frag(frag);
1358 	else
1359 		__free_page(virt_to_page(frag));
1360 }
1361 
1362 /**
1363  * nfp_net_rx_alloc_one() - Allocate and map page frag for RX
1364  * @dp:		NFP Net data path struct
1365  * @dma_addr:	Pointer to storage for DMA address (output param)
1366  *
1367  * This function will allcate a new page frag, map it for DMA.
1368  *
1369  * Return: allocated page frag or NULL on failure.
1370  */
1371 static void *nfp_net_rx_alloc_one(struct nfp_net_dp *dp, dma_addr_t *dma_addr)
1372 {
1373 	void *frag;
1374 
1375 	if (!dp->xdp_prog) {
1376 		frag = netdev_alloc_frag(dp->fl_bufsz);
1377 	} else {
1378 		struct page *page;
1379 
1380 		page = alloc_page(GFP_KERNEL);
1381 		frag = page ? page_address(page) : NULL;
1382 	}
1383 	if (!frag) {
1384 		nn_dp_warn(dp, "Failed to alloc receive page frag\n");
1385 		return NULL;
1386 	}
1387 
1388 	*dma_addr = nfp_net_dma_map_rx(dp, frag);
1389 	if (dma_mapping_error(dp->dev, *dma_addr)) {
1390 		nfp_net_free_frag(frag, dp->xdp_prog);
1391 		nn_dp_warn(dp, "Failed to map DMA RX buffer\n");
1392 		return NULL;
1393 	}
1394 
1395 	return frag;
1396 }
1397 
1398 static void *nfp_net_napi_alloc_one(struct nfp_net_dp *dp, dma_addr_t *dma_addr)
1399 {
1400 	void *frag;
1401 
1402 	if (!dp->xdp_prog) {
1403 		frag = napi_alloc_frag(dp->fl_bufsz);
1404 		if (unlikely(!frag))
1405 			return NULL;
1406 	} else {
1407 		struct page *page;
1408 
1409 		page = dev_alloc_page();
1410 		if (unlikely(!page))
1411 			return NULL;
1412 		frag = page_address(page);
1413 	}
1414 
1415 	*dma_addr = nfp_net_dma_map_rx(dp, frag);
1416 	if (dma_mapping_error(dp->dev, *dma_addr)) {
1417 		nfp_net_free_frag(frag, dp->xdp_prog);
1418 		nn_dp_warn(dp, "Failed to map DMA RX buffer\n");
1419 		return NULL;
1420 	}
1421 
1422 	return frag;
1423 }
1424 
1425 /**
1426  * nfp_net_rx_give_one() - Put mapped skb on the software and hardware rings
1427  * @dp:		NFP Net data path struct
1428  * @rx_ring:	RX ring structure
1429  * @frag:	page fragment buffer
1430  * @dma_addr:	DMA address of skb mapping
1431  */
1432 static void nfp_net_rx_give_one(const struct nfp_net_dp *dp,
1433 				struct nfp_net_rx_ring *rx_ring,
1434 				void *frag, dma_addr_t dma_addr)
1435 {
1436 	unsigned int wr_idx;
1437 
1438 	wr_idx = D_IDX(rx_ring, rx_ring->wr_p);
1439 
1440 	nfp_net_dma_sync_dev_rx(dp, dma_addr);
1441 
1442 	/* Stash SKB and DMA address away */
1443 	rx_ring->rxbufs[wr_idx].frag = frag;
1444 	rx_ring->rxbufs[wr_idx].dma_addr = dma_addr;
1445 
1446 	/* Fill freelist descriptor */
1447 	rx_ring->rxds[wr_idx].fld.reserved = 0;
1448 	rx_ring->rxds[wr_idx].fld.meta_len_dd = 0;
1449 	nfp_desc_set_dma_addr(&rx_ring->rxds[wr_idx].fld,
1450 			      dma_addr + dp->rx_dma_off);
1451 
1452 	rx_ring->wr_p++;
1453 	if (!(rx_ring->wr_p % NFP_NET_FL_BATCH)) {
1454 		/* Update write pointer of the freelist queue. Make
1455 		 * sure all writes are flushed before telling the hardware.
1456 		 */
1457 		wmb();
1458 		nfp_qcp_wr_ptr_add(rx_ring->qcp_fl, NFP_NET_FL_BATCH);
1459 	}
1460 }
1461 
1462 /**
1463  * nfp_net_rx_ring_reset() - Reflect in SW state of freelist after disable
1464  * @rx_ring:	RX ring structure
1465  *
1466  * Assumes that the device is stopped, must be idempotent.
1467  */
1468 static void nfp_net_rx_ring_reset(struct nfp_net_rx_ring *rx_ring)
1469 {
1470 	unsigned int wr_idx, last_idx;
1471 
1472 	/* wr_p == rd_p means ring was never fed FL bufs.  RX rings are always
1473 	 * kept at cnt - 1 FL bufs.
1474 	 */
1475 	if (rx_ring->wr_p == 0 && rx_ring->rd_p == 0)
1476 		return;
1477 
1478 	/* Move the empty entry to the end of the list */
1479 	wr_idx = D_IDX(rx_ring, rx_ring->wr_p);
1480 	last_idx = rx_ring->cnt - 1;
1481 	rx_ring->rxbufs[wr_idx].dma_addr = rx_ring->rxbufs[last_idx].dma_addr;
1482 	rx_ring->rxbufs[wr_idx].frag = rx_ring->rxbufs[last_idx].frag;
1483 	rx_ring->rxbufs[last_idx].dma_addr = 0;
1484 	rx_ring->rxbufs[last_idx].frag = NULL;
1485 
1486 	memset(rx_ring->rxds, 0, rx_ring->size);
1487 	rx_ring->wr_p = 0;
1488 	rx_ring->rd_p = 0;
1489 }
1490 
1491 /**
1492  * nfp_net_rx_ring_bufs_free() - Free any buffers currently on the RX ring
1493  * @dp:		NFP Net data path struct
1494  * @rx_ring:	RX ring to remove buffers from
1495  *
1496  * Assumes that the device is stopped and buffers are in [0, ring->cnt - 1)
1497  * entries.  After device is disabled nfp_net_rx_ring_reset() must be called
1498  * to restore required ring geometry.
1499  */
1500 static void
1501 nfp_net_rx_ring_bufs_free(struct nfp_net_dp *dp,
1502 			  struct nfp_net_rx_ring *rx_ring)
1503 {
1504 	unsigned int i;
1505 
1506 	for (i = 0; i < rx_ring->cnt - 1; i++) {
1507 		/* NULL skb can only happen when initial filling of the ring
1508 		 * fails to allocate enough buffers and calls here to free
1509 		 * already allocated ones.
1510 		 */
1511 		if (!rx_ring->rxbufs[i].frag)
1512 			continue;
1513 
1514 		nfp_net_dma_unmap_rx(dp, rx_ring->rxbufs[i].dma_addr);
1515 		nfp_net_free_frag(rx_ring->rxbufs[i].frag, dp->xdp_prog);
1516 		rx_ring->rxbufs[i].dma_addr = 0;
1517 		rx_ring->rxbufs[i].frag = NULL;
1518 	}
1519 }
1520 
1521 /**
1522  * nfp_net_rx_ring_bufs_alloc() - Fill RX ring with buffers (don't give to FW)
1523  * @dp:		NFP Net data path struct
1524  * @rx_ring:	RX ring to remove buffers from
1525  */
1526 static int
1527 nfp_net_rx_ring_bufs_alloc(struct nfp_net_dp *dp,
1528 			   struct nfp_net_rx_ring *rx_ring)
1529 {
1530 	struct nfp_net_rx_buf *rxbufs;
1531 	unsigned int i;
1532 
1533 	rxbufs = rx_ring->rxbufs;
1534 
1535 	for (i = 0; i < rx_ring->cnt - 1; i++) {
1536 		rxbufs[i].frag = nfp_net_rx_alloc_one(dp, &rxbufs[i].dma_addr);
1537 		if (!rxbufs[i].frag) {
1538 			nfp_net_rx_ring_bufs_free(dp, rx_ring);
1539 			return -ENOMEM;
1540 		}
1541 	}
1542 
1543 	return 0;
1544 }
1545 
1546 /**
1547  * nfp_net_rx_ring_fill_freelist() - Give buffers from the ring to FW
1548  * @dp:	     NFP Net data path struct
1549  * @rx_ring: RX ring to fill
1550  */
1551 static void
1552 nfp_net_rx_ring_fill_freelist(struct nfp_net_dp *dp,
1553 			      struct nfp_net_rx_ring *rx_ring)
1554 {
1555 	unsigned int i;
1556 
1557 	for (i = 0; i < rx_ring->cnt - 1; i++)
1558 		nfp_net_rx_give_one(dp, rx_ring, rx_ring->rxbufs[i].frag,
1559 				    rx_ring->rxbufs[i].dma_addr);
1560 }
1561 
1562 /**
1563  * nfp_net_rx_csum_has_errors() - group check if rxd has any csum errors
1564  * @flags: RX descriptor flags field in CPU byte order
1565  */
1566 static int nfp_net_rx_csum_has_errors(u16 flags)
1567 {
1568 	u16 csum_all_checked, csum_all_ok;
1569 
1570 	csum_all_checked = flags & __PCIE_DESC_RX_CSUM_ALL;
1571 	csum_all_ok = flags & __PCIE_DESC_RX_CSUM_ALL_OK;
1572 
1573 	return csum_all_checked != (csum_all_ok << PCIE_DESC_RX_CSUM_OK_SHIFT);
1574 }
1575 
1576 /**
1577  * nfp_net_rx_csum() - set SKB checksum field based on RX descriptor flags
1578  * @dp:  NFP Net data path struct
1579  * @r_vec: per-ring structure
1580  * @rxd: Pointer to RX descriptor
1581  * @meta: Parsed metadata prepend
1582  * @skb: Pointer to SKB
1583  */
1584 static void nfp_net_rx_csum(struct nfp_net_dp *dp,
1585 			    struct nfp_net_r_vector *r_vec,
1586 			    struct nfp_net_rx_desc *rxd,
1587 			    struct nfp_meta_parsed *meta, struct sk_buff *skb)
1588 {
1589 	skb_checksum_none_assert(skb);
1590 
1591 	if (!(dp->netdev->features & NETIF_F_RXCSUM))
1592 		return;
1593 
1594 	if (meta->csum_type) {
1595 		skb->ip_summed = meta->csum_type;
1596 		skb->csum = meta->csum;
1597 		u64_stats_update_begin(&r_vec->rx_sync);
1598 		r_vec->hw_csum_rx_complete++;
1599 		u64_stats_update_end(&r_vec->rx_sync);
1600 		return;
1601 	}
1602 
1603 	if (nfp_net_rx_csum_has_errors(le16_to_cpu(rxd->rxd.flags))) {
1604 		u64_stats_update_begin(&r_vec->rx_sync);
1605 		r_vec->hw_csum_rx_error++;
1606 		u64_stats_update_end(&r_vec->rx_sync);
1607 		return;
1608 	}
1609 
1610 	/* Assume that the firmware will never report inner CSUM_OK unless outer
1611 	 * L4 headers were successfully parsed. FW will always report zero UDP
1612 	 * checksum as CSUM_OK.
1613 	 */
1614 	if (rxd->rxd.flags & PCIE_DESC_RX_TCP_CSUM_OK ||
1615 	    rxd->rxd.flags & PCIE_DESC_RX_UDP_CSUM_OK) {
1616 		__skb_incr_checksum_unnecessary(skb);
1617 		u64_stats_update_begin(&r_vec->rx_sync);
1618 		r_vec->hw_csum_rx_ok++;
1619 		u64_stats_update_end(&r_vec->rx_sync);
1620 	}
1621 
1622 	if (rxd->rxd.flags & PCIE_DESC_RX_I_TCP_CSUM_OK ||
1623 	    rxd->rxd.flags & PCIE_DESC_RX_I_UDP_CSUM_OK) {
1624 		__skb_incr_checksum_unnecessary(skb);
1625 		u64_stats_update_begin(&r_vec->rx_sync);
1626 		r_vec->hw_csum_rx_inner_ok++;
1627 		u64_stats_update_end(&r_vec->rx_sync);
1628 	}
1629 }
1630 
1631 static void
1632 nfp_net_set_hash(struct net_device *netdev, struct nfp_meta_parsed *meta,
1633 		 unsigned int type, __be32 *hash)
1634 {
1635 	if (!(netdev->features & NETIF_F_RXHASH))
1636 		return;
1637 
1638 	switch (type) {
1639 	case NFP_NET_RSS_IPV4:
1640 	case NFP_NET_RSS_IPV6:
1641 	case NFP_NET_RSS_IPV6_EX:
1642 		meta->hash_type = PKT_HASH_TYPE_L3;
1643 		break;
1644 	default:
1645 		meta->hash_type = PKT_HASH_TYPE_L4;
1646 		break;
1647 	}
1648 
1649 	meta->hash = get_unaligned_be32(hash);
1650 }
1651 
1652 static void
1653 nfp_net_set_hash_desc(struct net_device *netdev, struct nfp_meta_parsed *meta,
1654 		      void *data, struct nfp_net_rx_desc *rxd)
1655 {
1656 	struct nfp_net_rx_hash *rx_hash = data;
1657 
1658 	if (!(rxd->rxd.flags & PCIE_DESC_RX_RSS))
1659 		return;
1660 
1661 	nfp_net_set_hash(netdev, meta, get_unaligned_be32(&rx_hash->hash_type),
1662 			 &rx_hash->hash);
1663 }
1664 
1665 static bool
1666 nfp_net_parse_meta(struct net_device *netdev, struct nfp_meta_parsed *meta,
1667 		   void *data, void *pkt, unsigned int pkt_len, int meta_len)
1668 {
1669 	u32 meta_info;
1670 
1671 	meta_info = get_unaligned_be32(data);
1672 	data += 4;
1673 
1674 	while (meta_info) {
1675 		switch (meta_info & NFP_NET_META_FIELD_MASK) {
1676 		case NFP_NET_META_HASH:
1677 			meta_info >>= NFP_NET_META_FIELD_SIZE;
1678 			nfp_net_set_hash(netdev, meta,
1679 					 meta_info & NFP_NET_META_FIELD_MASK,
1680 					 (__be32 *)data);
1681 			data += 4;
1682 			break;
1683 		case NFP_NET_META_MARK:
1684 			meta->mark = get_unaligned_be32(data);
1685 			data += 4;
1686 			break;
1687 		case NFP_NET_META_PORTID:
1688 			meta->portid = get_unaligned_be32(data);
1689 			data += 4;
1690 			break;
1691 		case NFP_NET_META_CSUM:
1692 			meta->csum_type = CHECKSUM_COMPLETE;
1693 			meta->csum =
1694 				(__force __wsum)__get_unaligned_cpu32(data);
1695 			data += 4;
1696 			break;
1697 		case NFP_NET_META_RESYNC_INFO:
1698 			if (nfp_net_tls_rx_resync_req(netdev, data, pkt,
1699 						      pkt_len))
1700 				return NULL;
1701 			data += sizeof(struct nfp_net_tls_resync_req);
1702 			break;
1703 		default:
1704 			return true;
1705 		}
1706 
1707 		meta_info >>= NFP_NET_META_FIELD_SIZE;
1708 	}
1709 
1710 	return data != pkt;
1711 }
1712 
1713 static void
1714 nfp_net_rx_drop(const struct nfp_net_dp *dp, struct nfp_net_r_vector *r_vec,
1715 		struct nfp_net_rx_ring *rx_ring, struct nfp_net_rx_buf *rxbuf,
1716 		struct sk_buff *skb)
1717 {
1718 	u64_stats_update_begin(&r_vec->rx_sync);
1719 	r_vec->rx_drops++;
1720 	/* If we have both skb and rxbuf the replacement buffer allocation
1721 	 * must have failed, count this as an alloc failure.
1722 	 */
1723 	if (skb && rxbuf)
1724 		r_vec->rx_replace_buf_alloc_fail++;
1725 	u64_stats_update_end(&r_vec->rx_sync);
1726 
1727 	/* skb is build based on the frag, free_skb() would free the frag
1728 	 * so to be able to reuse it we need an extra ref.
1729 	 */
1730 	if (skb && rxbuf && skb->head == rxbuf->frag)
1731 		page_ref_inc(virt_to_head_page(rxbuf->frag));
1732 	if (rxbuf)
1733 		nfp_net_rx_give_one(dp, rx_ring, rxbuf->frag, rxbuf->dma_addr);
1734 	if (skb)
1735 		dev_kfree_skb_any(skb);
1736 }
1737 
1738 static bool
1739 nfp_net_tx_xdp_buf(struct nfp_net_dp *dp, struct nfp_net_rx_ring *rx_ring,
1740 		   struct nfp_net_tx_ring *tx_ring,
1741 		   struct nfp_net_rx_buf *rxbuf, unsigned int dma_off,
1742 		   unsigned int pkt_len, bool *completed)
1743 {
1744 	unsigned int dma_map_sz = dp->fl_bufsz - NFP_NET_RX_BUF_NON_DATA;
1745 	struct nfp_net_tx_buf *txbuf;
1746 	struct nfp_net_tx_desc *txd;
1747 	int wr_idx;
1748 
1749 	/* Reject if xdp_adjust_tail grow packet beyond DMA area */
1750 	if (pkt_len + dma_off > dma_map_sz)
1751 		return false;
1752 
1753 	if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
1754 		if (!*completed) {
1755 			nfp_net_xdp_complete(tx_ring);
1756 			*completed = true;
1757 		}
1758 
1759 		if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
1760 			nfp_net_rx_drop(dp, rx_ring->r_vec, rx_ring, rxbuf,
1761 					NULL);
1762 			return false;
1763 		}
1764 	}
1765 
1766 	wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
1767 
1768 	/* Stash the soft descriptor of the head then initialize it */
1769 	txbuf = &tx_ring->txbufs[wr_idx];
1770 
1771 	nfp_net_rx_give_one(dp, rx_ring, txbuf->frag, txbuf->dma_addr);
1772 
1773 	txbuf->frag = rxbuf->frag;
1774 	txbuf->dma_addr = rxbuf->dma_addr;
1775 	txbuf->fidx = -1;
1776 	txbuf->pkt_cnt = 1;
1777 	txbuf->real_len = pkt_len;
1778 
1779 	dma_sync_single_for_device(dp->dev, rxbuf->dma_addr + dma_off,
1780 				   pkt_len, DMA_BIDIRECTIONAL);
1781 
1782 	/* Build TX descriptor */
1783 	txd = &tx_ring->txds[wr_idx];
1784 	txd->offset_eop = PCIE_DESC_TX_EOP;
1785 	txd->dma_len = cpu_to_le16(pkt_len);
1786 	nfp_desc_set_dma_addr(txd, rxbuf->dma_addr + dma_off);
1787 	txd->data_len = cpu_to_le16(pkt_len);
1788 
1789 	txd->flags = 0;
1790 	txd->mss = 0;
1791 	txd->lso_hdrlen = 0;
1792 
1793 	tx_ring->wr_p++;
1794 	tx_ring->wr_ptr_add++;
1795 	return true;
1796 }
1797 
1798 /**
1799  * nfp_net_rx() - receive up to @budget packets on @rx_ring
1800  * @rx_ring:   RX ring to receive from
1801  * @budget:    NAPI budget
1802  *
1803  * Note, this function is separated out from the napi poll function to
1804  * more cleanly separate packet receive code from other bookkeeping
1805  * functions performed in the napi poll function.
1806  *
1807  * Return: Number of packets received.
1808  */
1809 static int nfp_net_rx(struct nfp_net_rx_ring *rx_ring, int budget)
1810 {
1811 	struct nfp_net_r_vector *r_vec = rx_ring->r_vec;
1812 	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
1813 	struct nfp_net_tx_ring *tx_ring;
1814 	struct bpf_prog *xdp_prog;
1815 	bool xdp_tx_cmpl = false;
1816 	unsigned int true_bufsz;
1817 	struct sk_buff *skb;
1818 	int pkts_polled = 0;
1819 	struct xdp_buff xdp;
1820 	int idx;
1821 
1822 	rcu_read_lock();
1823 	xdp_prog = READ_ONCE(dp->xdp_prog);
1824 	true_bufsz = xdp_prog ? PAGE_SIZE : dp->fl_bufsz;
1825 	xdp_init_buff(&xdp, PAGE_SIZE - NFP_NET_RX_BUF_HEADROOM,
1826 		      &rx_ring->xdp_rxq);
1827 	tx_ring = r_vec->xdp_ring;
1828 
1829 	while (pkts_polled < budget) {
1830 		unsigned int meta_len, data_len, meta_off, pkt_len, pkt_off;
1831 		struct nfp_net_rx_buf *rxbuf;
1832 		struct nfp_net_rx_desc *rxd;
1833 		struct nfp_meta_parsed meta;
1834 		bool redir_egress = false;
1835 		struct net_device *netdev;
1836 		dma_addr_t new_dma_addr;
1837 		u32 meta_len_xdp = 0;
1838 		void *new_frag;
1839 
1840 		idx = D_IDX(rx_ring, rx_ring->rd_p);
1841 
1842 		rxd = &rx_ring->rxds[idx];
1843 		if (!(rxd->rxd.meta_len_dd & PCIE_DESC_RX_DD))
1844 			break;
1845 
1846 		/* Memory barrier to ensure that we won't do other reads
1847 		 * before the DD bit.
1848 		 */
1849 		dma_rmb();
1850 
1851 		memset(&meta, 0, sizeof(meta));
1852 
1853 		rx_ring->rd_p++;
1854 		pkts_polled++;
1855 
1856 		rxbuf =	&rx_ring->rxbufs[idx];
1857 		/*         < meta_len >
1858 		 *  <-- [rx_offset] -->
1859 		 *  ---------------------------------------------------------
1860 		 * | [XX] |  metadata  |             packet           | XXXX |
1861 		 *  ---------------------------------------------------------
1862 		 *         <---------------- data_len --------------->
1863 		 *
1864 		 * The rx_offset is fixed for all packets, the meta_len can vary
1865 		 * on a packet by packet basis. If rx_offset is set to zero
1866 		 * (_RX_OFFSET_DYNAMIC) metadata starts at the beginning of the
1867 		 * buffer and is immediately followed by the packet (no [XX]).
1868 		 */
1869 		meta_len = rxd->rxd.meta_len_dd & PCIE_DESC_RX_META_LEN_MASK;
1870 		data_len = le16_to_cpu(rxd->rxd.data_len);
1871 		pkt_len = data_len - meta_len;
1872 
1873 		pkt_off = NFP_NET_RX_BUF_HEADROOM + dp->rx_dma_off;
1874 		if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
1875 			pkt_off += meta_len;
1876 		else
1877 			pkt_off += dp->rx_offset;
1878 		meta_off = pkt_off - meta_len;
1879 
1880 		/* Stats update */
1881 		u64_stats_update_begin(&r_vec->rx_sync);
1882 		r_vec->rx_pkts++;
1883 		r_vec->rx_bytes += pkt_len;
1884 		u64_stats_update_end(&r_vec->rx_sync);
1885 
1886 		if (unlikely(meta_len > NFP_NET_MAX_PREPEND ||
1887 			     (dp->rx_offset && meta_len > dp->rx_offset))) {
1888 			nn_dp_warn(dp, "oversized RX packet metadata %u\n",
1889 				   meta_len);
1890 			nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
1891 			continue;
1892 		}
1893 
1894 		nfp_net_dma_sync_cpu_rx(dp, rxbuf->dma_addr + meta_off,
1895 					data_len);
1896 
1897 		if (!dp->chained_metadata_format) {
1898 			nfp_net_set_hash_desc(dp->netdev, &meta,
1899 					      rxbuf->frag + meta_off, rxd);
1900 		} else if (meta_len) {
1901 			if (unlikely(nfp_net_parse_meta(dp->netdev, &meta,
1902 							rxbuf->frag + meta_off,
1903 							rxbuf->frag + pkt_off,
1904 							pkt_len, meta_len))) {
1905 				nn_dp_warn(dp, "invalid RX packet metadata\n");
1906 				nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf,
1907 						NULL);
1908 				continue;
1909 			}
1910 		}
1911 
1912 		if (xdp_prog && !meta.portid) {
1913 			void *orig_data = rxbuf->frag + pkt_off;
1914 			unsigned int dma_off;
1915 			int act;
1916 
1917 			xdp_prepare_buff(&xdp,
1918 					 rxbuf->frag + NFP_NET_RX_BUF_HEADROOM,
1919 					 pkt_off - NFP_NET_RX_BUF_HEADROOM,
1920 					 pkt_len, true);
1921 
1922 			act = bpf_prog_run_xdp(xdp_prog, &xdp);
1923 
1924 			pkt_len = xdp.data_end - xdp.data;
1925 			pkt_off += xdp.data - orig_data;
1926 
1927 			switch (act) {
1928 			case XDP_PASS:
1929 				meta_len_xdp = xdp.data - xdp.data_meta;
1930 				break;
1931 			case XDP_TX:
1932 				dma_off = pkt_off - NFP_NET_RX_BUF_HEADROOM;
1933 				if (unlikely(!nfp_net_tx_xdp_buf(dp, rx_ring,
1934 								 tx_ring, rxbuf,
1935 								 dma_off,
1936 								 pkt_len,
1937 								 &xdp_tx_cmpl)))
1938 					trace_xdp_exception(dp->netdev,
1939 							    xdp_prog, act);
1940 				continue;
1941 			default:
1942 				bpf_warn_invalid_xdp_action(act);
1943 				fallthrough;
1944 			case XDP_ABORTED:
1945 				trace_xdp_exception(dp->netdev, xdp_prog, act);
1946 				fallthrough;
1947 			case XDP_DROP:
1948 				nfp_net_rx_give_one(dp, rx_ring, rxbuf->frag,
1949 						    rxbuf->dma_addr);
1950 				continue;
1951 			}
1952 		}
1953 
1954 		if (likely(!meta.portid)) {
1955 			netdev = dp->netdev;
1956 		} else if (meta.portid == NFP_META_PORT_ID_CTRL) {
1957 			struct nfp_net *nn = netdev_priv(dp->netdev);
1958 
1959 			nfp_app_ctrl_rx_raw(nn->app, rxbuf->frag + pkt_off,
1960 					    pkt_len);
1961 			nfp_net_rx_give_one(dp, rx_ring, rxbuf->frag,
1962 					    rxbuf->dma_addr);
1963 			continue;
1964 		} else {
1965 			struct nfp_net *nn;
1966 
1967 			nn = netdev_priv(dp->netdev);
1968 			netdev = nfp_app_dev_get(nn->app, meta.portid,
1969 						 &redir_egress);
1970 			if (unlikely(!netdev)) {
1971 				nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf,
1972 						NULL);
1973 				continue;
1974 			}
1975 
1976 			if (nfp_netdev_is_nfp_repr(netdev))
1977 				nfp_repr_inc_rx_stats(netdev, pkt_len);
1978 		}
1979 
1980 		skb = build_skb(rxbuf->frag, true_bufsz);
1981 		if (unlikely(!skb)) {
1982 			nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
1983 			continue;
1984 		}
1985 		new_frag = nfp_net_napi_alloc_one(dp, &new_dma_addr);
1986 		if (unlikely(!new_frag)) {
1987 			nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, skb);
1988 			continue;
1989 		}
1990 
1991 		nfp_net_dma_unmap_rx(dp, rxbuf->dma_addr);
1992 
1993 		nfp_net_rx_give_one(dp, rx_ring, new_frag, new_dma_addr);
1994 
1995 		skb_reserve(skb, pkt_off);
1996 		skb_put(skb, pkt_len);
1997 
1998 		skb->mark = meta.mark;
1999 		skb_set_hash(skb, meta.hash, meta.hash_type);
2000 
2001 		skb_record_rx_queue(skb, rx_ring->idx);
2002 		skb->protocol = eth_type_trans(skb, netdev);
2003 
2004 		nfp_net_rx_csum(dp, r_vec, rxd, &meta, skb);
2005 
2006 #ifdef CONFIG_TLS_DEVICE
2007 		if (rxd->rxd.flags & PCIE_DESC_RX_DECRYPTED) {
2008 			skb->decrypted = true;
2009 			u64_stats_update_begin(&r_vec->rx_sync);
2010 			r_vec->hw_tls_rx++;
2011 			u64_stats_update_end(&r_vec->rx_sync);
2012 		}
2013 #endif
2014 
2015 		if (rxd->rxd.flags & PCIE_DESC_RX_VLAN)
2016 			__vlan_hwaccel_put_tag(skb, htons(ETH_P_8021Q),
2017 					       le16_to_cpu(rxd->rxd.vlan));
2018 		if (meta_len_xdp)
2019 			skb_metadata_set(skb, meta_len_xdp);
2020 
2021 		if (likely(!redir_egress)) {
2022 			napi_gro_receive(&rx_ring->r_vec->napi, skb);
2023 		} else {
2024 			skb->dev = netdev;
2025 			skb_reset_network_header(skb);
2026 			__skb_push(skb, ETH_HLEN);
2027 			dev_queue_xmit(skb);
2028 		}
2029 	}
2030 
2031 	if (xdp_prog) {
2032 		if (tx_ring->wr_ptr_add)
2033 			nfp_net_tx_xmit_more_flush(tx_ring);
2034 		else if (unlikely(tx_ring->wr_p != tx_ring->rd_p) &&
2035 			 !xdp_tx_cmpl)
2036 			if (!nfp_net_xdp_complete(tx_ring))
2037 				pkts_polled = budget;
2038 	}
2039 	rcu_read_unlock();
2040 
2041 	return pkts_polled;
2042 }
2043 
2044 /**
2045  * nfp_net_poll() - napi poll function
2046  * @napi:    NAPI structure
2047  * @budget:  NAPI budget
2048  *
2049  * Return: number of packets polled.
2050  */
2051 static int nfp_net_poll(struct napi_struct *napi, int budget)
2052 {
2053 	struct nfp_net_r_vector *r_vec =
2054 		container_of(napi, struct nfp_net_r_vector, napi);
2055 	unsigned int pkts_polled = 0;
2056 
2057 	if (r_vec->tx_ring)
2058 		nfp_net_tx_complete(r_vec->tx_ring, budget);
2059 	if (r_vec->rx_ring)
2060 		pkts_polled = nfp_net_rx(r_vec->rx_ring, budget);
2061 
2062 	if (pkts_polled < budget)
2063 		if (napi_complete_done(napi, pkts_polled))
2064 			nfp_net_irq_unmask(r_vec->nfp_net, r_vec->irq_entry);
2065 
2066 	return pkts_polled;
2067 }
2068 
2069 /* Control device data path
2070  */
2071 
2072 static bool
2073 nfp_ctrl_tx_one(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
2074 		struct sk_buff *skb, bool old)
2075 {
2076 	unsigned int real_len = skb->len, meta_len = 0;
2077 	struct nfp_net_tx_ring *tx_ring;
2078 	struct nfp_net_tx_buf *txbuf;
2079 	struct nfp_net_tx_desc *txd;
2080 	struct nfp_net_dp *dp;
2081 	dma_addr_t dma_addr;
2082 	int wr_idx;
2083 
2084 	dp = &r_vec->nfp_net->dp;
2085 	tx_ring = r_vec->tx_ring;
2086 
2087 	if (WARN_ON_ONCE(skb_shinfo(skb)->nr_frags)) {
2088 		nn_dp_warn(dp, "Driver's CTRL TX does not implement gather\n");
2089 		goto err_free;
2090 	}
2091 
2092 	if (unlikely(nfp_net_tx_full(tx_ring, 1))) {
2093 		u64_stats_update_begin(&r_vec->tx_sync);
2094 		r_vec->tx_busy++;
2095 		u64_stats_update_end(&r_vec->tx_sync);
2096 		if (!old)
2097 			__skb_queue_tail(&r_vec->queue, skb);
2098 		else
2099 			__skb_queue_head(&r_vec->queue, skb);
2100 		return true;
2101 	}
2102 
2103 	if (nfp_app_ctrl_has_meta(nn->app)) {
2104 		if (unlikely(skb_headroom(skb) < 8)) {
2105 			nn_dp_warn(dp, "CTRL TX on skb without headroom\n");
2106 			goto err_free;
2107 		}
2108 		meta_len = 8;
2109 		put_unaligned_be32(NFP_META_PORT_ID_CTRL, skb_push(skb, 4));
2110 		put_unaligned_be32(NFP_NET_META_PORTID, skb_push(skb, 4));
2111 	}
2112 
2113 	/* Start with the head skbuf */
2114 	dma_addr = dma_map_single(dp->dev, skb->data, skb_headlen(skb),
2115 				  DMA_TO_DEVICE);
2116 	if (dma_mapping_error(dp->dev, dma_addr))
2117 		goto err_dma_warn;
2118 
2119 	wr_idx = D_IDX(tx_ring, tx_ring->wr_p);
2120 
2121 	/* Stash the soft descriptor of the head then initialize it */
2122 	txbuf = &tx_ring->txbufs[wr_idx];
2123 	txbuf->skb = skb;
2124 	txbuf->dma_addr = dma_addr;
2125 	txbuf->fidx = -1;
2126 	txbuf->pkt_cnt = 1;
2127 	txbuf->real_len = real_len;
2128 
2129 	/* Build TX descriptor */
2130 	txd = &tx_ring->txds[wr_idx];
2131 	txd->offset_eop = meta_len | PCIE_DESC_TX_EOP;
2132 	txd->dma_len = cpu_to_le16(skb_headlen(skb));
2133 	nfp_desc_set_dma_addr(txd, dma_addr);
2134 	txd->data_len = cpu_to_le16(skb->len);
2135 
2136 	txd->flags = 0;
2137 	txd->mss = 0;
2138 	txd->lso_hdrlen = 0;
2139 
2140 	tx_ring->wr_p++;
2141 	tx_ring->wr_ptr_add++;
2142 	nfp_net_tx_xmit_more_flush(tx_ring);
2143 
2144 	return false;
2145 
2146 err_dma_warn:
2147 	nn_dp_warn(dp, "Failed to DMA map TX CTRL buffer\n");
2148 err_free:
2149 	u64_stats_update_begin(&r_vec->tx_sync);
2150 	r_vec->tx_errors++;
2151 	u64_stats_update_end(&r_vec->tx_sync);
2152 	dev_kfree_skb_any(skb);
2153 	return false;
2154 }
2155 
2156 bool __nfp_ctrl_tx(struct nfp_net *nn, struct sk_buff *skb)
2157 {
2158 	struct nfp_net_r_vector *r_vec = &nn->r_vecs[0];
2159 
2160 	return nfp_ctrl_tx_one(nn, r_vec, skb, false);
2161 }
2162 
2163 bool nfp_ctrl_tx(struct nfp_net *nn, struct sk_buff *skb)
2164 {
2165 	struct nfp_net_r_vector *r_vec = &nn->r_vecs[0];
2166 	bool ret;
2167 
2168 	spin_lock_bh(&r_vec->lock);
2169 	ret = nfp_ctrl_tx_one(nn, r_vec, skb, false);
2170 	spin_unlock_bh(&r_vec->lock);
2171 
2172 	return ret;
2173 }
2174 
2175 static void __nfp_ctrl_tx_queued(struct nfp_net_r_vector *r_vec)
2176 {
2177 	struct sk_buff *skb;
2178 
2179 	while ((skb = __skb_dequeue(&r_vec->queue)))
2180 		if (nfp_ctrl_tx_one(r_vec->nfp_net, r_vec, skb, true))
2181 			return;
2182 }
2183 
2184 static bool
2185 nfp_ctrl_meta_ok(struct nfp_net *nn, void *data, unsigned int meta_len)
2186 {
2187 	u32 meta_type, meta_tag;
2188 
2189 	if (!nfp_app_ctrl_has_meta(nn->app))
2190 		return !meta_len;
2191 
2192 	if (meta_len != 8)
2193 		return false;
2194 
2195 	meta_type = get_unaligned_be32(data);
2196 	meta_tag = get_unaligned_be32(data + 4);
2197 
2198 	return (meta_type == NFP_NET_META_PORTID &&
2199 		meta_tag == NFP_META_PORT_ID_CTRL);
2200 }
2201 
2202 static bool
2203 nfp_ctrl_rx_one(struct nfp_net *nn, struct nfp_net_dp *dp,
2204 		struct nfp_net_r_vector *r_vec, struct nfp_net_rx_ring *rx_ring)
2205 {
2206 	unsigned int meta_len, data_len, meta_off, pkt_len, pkt_off;
2207 	struct nfp_net_rx_buf *rxbuf;
2208 	struct nfp_net_rx_desc *rxd;
2209 	dma_addr_t new_dma_addr;
2210 	struct sk_buff *skb;
2211 	void *new_frag;
2212 	int idx;
2213 
2214 	idx = D_IDX(rx_ring, rx_ring->rd_p);
2215 
2216 	rxd = &rx_ring->rxds[idx];
2217 	if (!(rxd->rxd.meta_len_dd & PCIE_DESC_RX_DD))
2218 		return false;
2219 
2220 	/* Memory barrier to ensure that we won't do other reads
2221 	 * before the DD bit.
2222 	 */
2223 	dma_rmb();
2224 
2225 	rx_ring->rd_p++;
2226 
2227 	rxbuf =	&rx_ring->rxbufs[idx];
2228 	meta_len = rxd->rxd.meta_len_dd & PCIE_DESC_RX_META_LEN_MASK;
2229 	data_len = le16_to_cpu(rxd->rxd.data_len);
2230 	pkt_len = data_len - meta_len;
2231 
2232 	pkt_off = NFP_NET_RX_BUF_HEADROOM + dp->rx_dma_off;
2233 	if (dp->rx_offset == NFP_NET_CFG_RX_OFFSET_DYNAMIC)
2234 		pkt_off += meta_len;
2235 	else
2236 		pkt_off += dp->rx_offset;
2237 	meta_off = pkt_off - meta_len;
2238 
2239 	/* Stats update */
2240 	u64_stats_update_begin(&r_vec->rx_sync);
2241 	r_vec->rx_pkts++;
2242 	r_vec->rx_bytes += pkt_len;
2243 	u64_stats_update_end(&r_vec->rx_sync);
2244 
2245 	nfp_net_dma_sync_cpu_rx(dp, rxbuf->dma_addr + meta_off,	data_len);
2246 
2247 	if (unlikely(!nfp_ctrl_meta_ok(nn, rxbuf->frag + meta_off, meta_len))) {
2248 		nn_dp_warn(dp, "incorrect metadata for ctrl packet (%d)\n",
2249 			   meta_len);
2250 		nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
2251 		return true;
2252 	}
2253 
2254 	skb = build_skb(rxbuf->frag, dp->fl_bufsz);
2255 	if (unlikely(!skb)) {
2256 		nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, NULL);
2257 		return true;
2258 	}
2259 	new_frag = nfp_net_napi_alloc_one(dp, &new_dma_addr);
2260 	if (unlikely(!new_frag)) {
2261 		nfp_net_rx_drop(dp, r_vec, rx_ring, rxbuf, skb);
2262 		return true;
2263 	}
2264 
2265 	nfp_net_dma_unmap_rx(dp, rxbuf->dma_addr);
2266 
2267 	nfp_net_rx_give_one(dp, rx_ring, new_frag, new_dma_addr);
2268 
2269 	skb_reserve(skb, pkt_off);
2270 	skb_put(skb, pkt_len);
2271 
2272 	nfp_app_ctrl_rx(nn->app, skb);
2273 
2274 	return true;
2275 }
2276 
2277 static bool nfp_ctrl_rx(struct nfp_net_r_vector *r_vec)
2278 {
2279 	struct nfp_net_rx_ring *rx_ring = r_vec->rx_ring;
2280 	struct nfp_net *nn = r_vec->nfp_net;
2281 	struct nfp_net_dp *dp = &nn->dp;
2282 	unsigned int budget = 512;
2283 
2284 	while (nfp_ctrl_rx_one(nn, dp, r_vec, rx_ring) && budget--)
2285 		continue;
2286 
2287 	return budget;
2288 }
2289 
2290 static void nfp_ctrl_poll(struct tasklet_struct *t)
2291 {
2292 	struct nfp_net_r_vector *r_vec = from_tasklet(r_vec, t, tasklet);
2293 
2294 	spin_lock(&r_vec->lock);
2295 	nfp_net_tx_complete(r_vec->tx_ring, 0);
2296 	__nfp_ctrl_tx_queued(r_vec);
2297 	spin_unlock(&r_vec->lock);
2298 
2299 	if (nfp_ctrl_rx(r_vec)) {
2300 		nfp_net_irq_unmask(r_vec->nfp_net, r_vec->irq_entry);
2301 	} else {
2302 		tasklet_schedule(&r_vec->tasklet);
2303 		nn_dp_warn(&r_vec->nfp_net->dp,
2304 			   "control message budget exceeded!\n");
2305 	}
2306 }
2307 
2308 /* Setup and Configuration
2309  */
2310 
2311 /**
2312  * nfp_net_vecs_init() - Assign IRQs and setup rvecs.
2313  * @nn:		NFP Network structure
2314  */
2315 static void nfp_net_vecs_init(struct nfp_net *nn)
2316 {
2317 	struct nfp_net_r_vector *r_vec;
2318 	int r;
2319 
2320 	nn->lsc_handler = nfp_net_irq_lsc;
2321 	nn->exn_handler = nfp_net_irq_exn;
2322 
2323 	for (r = 0; r < nn->max_r_vecs; r++) {
2324 		struct msix_entry *entry;
2325 
2326 		entry = &nn->irq_entries[NFP_NET_NON_Q_VECTORS + r];
2327 
2328 		r_vec = &nn->r_vecs[r];
2329 		r_vec->nfp_net = nn;
2330 		r_vec->irq_entry = entry->entry;
2331 		r_vec->irq_vector = entry->vector;
2332 
2333 		if (nn->dp.netdev) {
2334 			r_vec->handler = nfp_net_irq_rxtx;
2335 		} else {
2336 			r_vec->handler = nfp_ctrl_irq_rxtx;
2337 
2338 			__skb_queue_head_init(&r_vec->queue);
2339 			spin_lock_init(&r_vec->lock);
2340 			tasklet_setup(&r_vec->tasklet, nfp_ctrl_poll);
2341 			tasklet_disable(&r_vec->tasklet);
2342 		}
2343 
2344 		cpumask_set_cpu(r, &r_vec->affinity_mask);
2345 	}
2346 }
2347 
2348 /**
2349  * nfp_net_tx_ring_free() - Free resources allocated to a TX ring
2350  * @tx_ring:   TX ring to free
2351  */
2352 static void nfp_net_tx_ring_free(struct nfp_net_tx_ring *tx_ring)
2353 {
2354 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
2355 	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
2356 
2357 	kvfree(tx_ring->txbufs);
2358 
2359 	if (tx_ring->txds)
2360 		dma_free_coherent(dp->dev, tx_ring->size,
2361 				  tx_ring->txds, tx_ring->dma);
2362 
2363 	tx_ring->cnt = 0;
2364 	tx_ring->txbufs = NULL;
2365 	tx_ring->txds = NULL;
2366 	tx_ring->dma = 0;
2367 	tx_ring->size = 0;
2368 }
2369 
2370 /**
2371  * nfp_net_tx_ring_alloc() - Allocate resource for a TX ring
2372  * @dp:        NFP Net data path struct
2373  * @tx_ring:   TX Ring structure to allocate
2374  *
2375  * Return: 0 on success, negative errno otherwise.
2376  */
2377 static int
2378 nfp_net_tx_ring_alloc(struct nfp_net_dp *dp, struct nfp_net_tx_ring *tx_ring)
2379 {
2380 	struct nfp_net_r_vector *r_vec = tx_ring->r_vec;
2381 
2382 	tx_ring->cnt = dp->txd_cnt;
2383 
2384 	tx_ring->size = array_size(tx_ring->cnt, sizeof(*tx_ring->txds));
2385 	tx_ring->txds = dma_alloc_coherent(dp->dev, tx_ring->size,
2386 					   &tx_ring->dma,
2387 					   GFP_KERNEL | __GFP_NOWARN);
2388 	if (!tx_ring->txds) {
2389 		netdev_warn(dp->netdev, "failed to allocate TX descriptor ring memory, requested descriptor count: %d, consider lowering descriptor count\n",
2390 			    tx_ring->cnt);
2391 		goto err_alloc;
2392 	}
2393 
2394 	tx_ring->txbufs = kvcalloc(tx_ring->cnt, sizeof(*tx_ring->txbufs),
2395 				   GFP_KERNEL);
2396 	if (!tx_ring->txbufs)
2397 		goto err_alloc;
2398 
2399 	if (!tx_ring->is_xdp && dp->netdev)
2400 		netif_set_xps_queue(dp->netdev, &r_vec->affinity_mask,
2401 				    tx_ring->idx);
2402 
2403 	return 0;
2404 
2405 err_alloc:
2406 	nfp_net_tx_ring_free(tx_ring);
2407 	return -ENOMEM;
2408 }
2409 
2410 static void
2411 nfp_net_tx_ring_bufs_free(struct nfp_net_dp *dp,
2412 			  struct nfp_net_tx_ring *tx_ring)
2413 {
2414 	unsigned int i;
2415 
2416 	if (!tx_ring->is_xdp)
2417 		return;
2418 
2419 	for (i = 0; i < tx_ring->cnt; i++) {
2420 		if (!tx_ring->txbufs[i].frag)
2421 			return;
2422 
2423 		nfp_net_dma_unmap_rx(dp, tx_ring->txbufs[i].dma_addr);
2424 		__free_page(virt_to_page(tx_ring->txbufs[i].frag));
2425 	}
2426 }
2427 
2428 static int
2429 nfp_net_tx_ring_bufs_alloc(struct nfp_net_dp *dp,
2430 			   struct nfp_net_tx_ring *tx_ring)
2431 {
2432 	struct nfp_net_tx_buf *txbufs = tx_ring->txbufs;
2433 	unsigned int i;
2434 
2435 	if (!tx_ring->is_xdp)
2436 		return 0;
2437 
2438 	for (i = 0; i < tx_ring->cnt; i++) {
2439 		txbufs[i].frag = nfp_net_rx_alloc_one(dp, &txbufs[i].dma_addr);
2440 		if (!txbufs[i].frag) {
2441 			nfp_net_tx_ring_bufs_free(dp, tx_ring);
2442 			return -ENOMEM;
2443 		}
2444 	}
2445 
2446 	return 0;
2447 }
2448 
2449 static int nfp_net_tx_rings_prepare(struct nfp_net *nn, struct nfp_net_dp *dp)
2450 {
2451 	unsigned int r;
2452 
2453 	dp->tx_rings = kcalloc(dp->num_tx_rings, sizeof(*dp->tx_rings),
2454 			       GFP_KERNEL);
2455 	if (!dp->tx_rings)
2456 		return -ENOMEM;
2457 
2458 	for (r = 0; r < dp->num_tx_rings; r++) {
2459 		int bias = 0;
2460 
2461 		if (r >= dp->num_stack_tx_rings)
2462 			bias = dp->num_stack_tx_rings;
2463 
2464 		nfp_net_tx_ring_init(&dp->tx_rings[r], &nn->r_vecs[r - bias],
2465 				     r, bias);
2466 
2467 		if (nfp_net_tx_ring_alloc(dp, &dp->tx_rings[r]))
2468 			goto err_free_prev;
2469 
2470 		if (nfp_net_tx_ring_bufs_alloc(dp, &dp->tx_rings[r]))
2471 			goto err_free_ring;
2472 	}
2473 
2474 	return 0;
2475 
2476 err_free_prev:
2477 	while (r--) {
2478 		nfp_net_tx_ring_bufs_free(dp, &dp->tx_rings[r]);
2479 err_free_ring:
2480 		nfp_net_tx_ring_free(&dp->tx_rings[r]);
2481 	}
2482 	kfree(dp->tx_rings);
2483 	return -ENOMEM;
2484 }
2485 
2486 static void nfp_net_tx_rings_free(struct nfp_net_dp *dp)
2487 {
2488 	unsigned int r;
2489 
2490 	for (r = 0; r < dp->num_tx_rings; r++) {
2491 		nfp_net_tx_ring_bufs_free(dp, &dp->tx_rings[r]);
2492 		nfp_net_tx_ring_free(&dp->tx_rings[r]);
2493 	}
2494 
2495 	kfree(dp->tx_rings);
2496 }
2497 
2498 /**
2499  * nfp_net_rx_ring_free() - Free resources allocated to a RX ring
2500  * @rx_ring:  RX ring to free
2501  */
2502 static void nfp_net_rx_ring_free(struct nfp_net_rx_ring *rx_ring)
2503 {
2504 	struct nfp_net_r_vector *r_vec = rx_ring->r_vec;
2505 	struct nfp_net_dp *dp = &r_vec->nfp_net->dp;
2506 
2507 	if (dp->netdev)
2508 		xdp_rxq_info_unreg(&rx_ring->xdp_rxq);
2509 	kvfree(rx_ring->rxbufs);
2510 
2511 	if (rx_ring->rxds)
2512 		dma_free_coherent(dp->dev, rx_ring->size,
2513 				  rx_ring->rxds, rx_ring->dma);
2514 
2515 	rx_ring->cnt = 0;
2516 	rx_ring->rxbufs = NULL;
2517 	rx_ring->rxds = NULL;
2518 	rx_ring->dma = 0;
2519 	rx_ring->size = 0;
2520 }
2521 
2522 /**
2523  * nfp_net_rx_ring_alloc() - Allocate resource for a RX ring
2524  * @dp:	      NFP Net data path struct
2525  * @rx_ring:  RX ring to allocate
2526  *
2527  * Return: 0 on success, negative errno otherwise.
2528  */
2529 static int
2530 nfp_net_rx_ring_alloc(struct nfp_net_dp *dp, struct nfp_net_rx_ring *rx_ring)
2531 {
2532 	int err;
2533 
2534 	if (dp->netdev) {
2535 		err = xdp_rxq_info_reg(&rx_ring->xdp_rxq, dp->netdev,
2536 				       rx_ring->idx, rx_ring->r_vec->napi.napi_id);
2537 		if (err < 0)
2538 			return err;
2539 	}
2540 
2541 	rx_ring->cnt = dp->rxd_cnt;
2542 	rx_ring->size = array_size(rx_ring->cnt, sizeof(*rx_ring->rxds));
2543 	rx_ring->rxds = dma_alloc_coherent(dp->dev, rx_ring->size,
2544 					   &rx_ring->dma,
2545 					   GFP_KERNEL | __GFP_NOWARN);
2546 	if (!rx_ring->rxds) {
2547 		netdev_warn(dp->netdev, "failed to allocate RX descriptor ring memory, requested descriptor count: %d, consider lowering descriptor count\n",
2548 			    rx_ring->cnt);
2549 		goto err_alloc;
2550 	}
2551 
2552 	rx_ring->rxbufs = kvcalloc(rx_ring->cnt, sizeof(*rx_ring->rxbufs),
2553 				   GFP_KERNEL);
2554 	if (!rx_ring->rxbufs)
2555 		goto err_alloc;
2556 
2557 	return 0;
2558 
2559 err_alloc:
2560 	nfp_net_rx_ring_free(rx_ring);
2561 	return -ENOMEM;
2562 }
2563 
2564 static int nfp_net_rx_rings_prepare(struct nfp_net *nn, struct nfp_net_dp *dp)
2565 {
2566 	unsigned int r;
2567 
2568 	dp->rx_rings = kcalloc(dp->num_rx_rings, sizeof(*dp->rx_rings),
2569 			       GFP_KERNEL);
2570 	if (!dp->rx_rings)
2571 		return -ENOMEM;
2572 
2573 	for (r = 0; r < dp->num_rx_rings; r++) {
2574 		nfp_net_rx_ring_init(&dp->rx_rings[r], &nn->r_vecs[r], r);
2575 
2576 		if (nfp_net_rx_ring_alloc(dp, &dp->rx_rings[r]))
2577 			goto err_free_prev;
2578 
2579 		if (nfp_net_rx_ring_bufs_alloc(dp, &dp->rx_rings[r]))
2580 			goto err_free_ring;
2581 	}
2582 
2583 	return 0;
2584 
2585 err_free_prev:
2586 	while (r--) {
2587 		nfp_net_rx_ring_bufs_free(dp, &dp->rx_rings[r]);
2588 err_free_ring:
2589 		nfp_net_rx_ring_free(&dp->rx_rings[r]);
2590 	}
2591 	kfree(dp->rx_rings);
2592 	return -ENOMEM;
2593 }
2594 
2595 static void nfp_net_rx_rings_free(struct nfp_net_dp *dp)
2596 {
2597 	unsigned int r;
2598 
2599 	for (r = 0; r < dp->num_rx_rings; r++) {
2600 		nfp_net_rx_ring_bufs_free(dp, &dp->rx_rings[r]);
2601 		nfp_net_rx_ring_free(&dp->rx_rings[r]);
2602 	}
2603 
2604 	kfree(dp->rx_rings);
2605 }
2606 
2607 static void
2608 nfp_net_vector_assign_rings(struct nfp_net_dp *dp,
2609 			    struct nfp_net_r_vector *r_vec, int idx)
2610 {
2611 	r_vec->rx_ring = idx < dp->num_rx_rings ? &dp->rx_rings[idx] : NULL;
2612 	r_vec->tx_ring =
2613 		idx < dp->num_stack_tx_rings ? &dp->tx_rings[idx] : NULL;
2614 
2615 	r_vec->xdp_ring = idx < dp->num_tx_rings - dp->num_stack_tx_rings ?
2616 		&dp->tx_rings[dp->num_stack_tx_rings + idx] : NULL;
2617 }
2618 
2619 static int
2620 nfp_net_prepare_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec,
2621 		       int idx)
2622 {
2623 	int err;
2624 
2625 	/* Setup NAPI */
2626 	if (nn->dp.netdev)
2627 		netif_napi_add(nn->dp.netdev, &r_vec->napi,
2628 			       nfp_net_poll, NAPI_POLL_WEIGHT);
2629 	else
2630 		tasklet_enable(&r_vec->tasklet);
2631 
2632 	snprintf(r_vec->name, sizeof(r_vec->name),
2633 		 "%s-rxtx-%d", nfp_net_name(nn), idx);
2634 	err = request_irq(r_vec->irq_vector, r_vec->handler, 0, r_vec->name,
2635 			  r_vec);
2636 	if (err) {
2637 		if (nn->dp.netdev)
2638 			netif_napi_del(&r_vec->napi);
2639 		else
2640 			tasklet_disable(&r_vec->tasklet);
2641 
2642 		nn_err(nn, "Error requesting IRQ %d\n", r_vec->irq_vector);
2643 		return err;
2644 	}
2645 	disable_irq(r_vec->irq_vector);
2646 
2647 	irq_set_affinity_hint(r_vec->irq_vector, &r_vec->affinity_mask);
2648 
2649 	nn_dbg(nn, "RV%02d: irq=%03d/%03d\n", idx, r_vec->irq_vector,
2650 	       r_vec->irq_entry);
2651 
2652 	return 0;
2653 }
2654 
2655 static void
2656 nfp_net_cleanup_vector(struct nfp_net *nn, struct nfp_net_r_vector *r_vec)
2657 {
2658 	irq_set_affinity_hint(r_vec->irq_vector, NULL);
2659 	if (nn->dp.netdev)
2660 		netif_napi_del(&r_vec->napi);
2661 	else
2662 		tasklet_disable(&r_vec->tasklet);
2663 
2664 	free_irq(r_vec->irq_vector, r_vec);
2665 }
2666 
2667 /**
2668  * nfp_net_rss_write_itbl() - Write RSS indirection table to device
2669  * @nn:      NFP Net device to reconfigure
2670  */
2671 void nfp_net_rss_write_itbl(struct nfp_net *nn)
2672 {
2673 	int i;
2674 
2675 	for (i = 0; i < NFP_NET_CFG_RSS_ITBL_SZ; i += 4)
2676 		nn_writel(nn, NFP_NET_CFG_RSS_ITBL + i,
2677 			  get_unaligned_le32(nn->rss_itbl + i));
2678 }
2679 
2680 /**
2681  * nfp_net_rss_write_key() - Write RSS hash key to device
2682  * @nn:      NFP Net device to reconfigure
2683  */
2684 void nfp_net_rss_write_key(struct nfp_net *nn)
2685 {
2686 	int i;
2687 
2688 	for (i = 0; i < nfp_net_rss_key_sz(nn); i += 4)
2689 		nn_writel(nn, NFP_NET_CFG_RSS_KEY + i,
2690 			  get_unaligned_le32(nn->rss_key + i));
2691 }
2692 
2693 /**
2694  * nfp_net_coalesce_write_cfg() - Write irq coalescence configuration to HW
2695  * @nn:      NFP Net device to reconfigure
2696  */
2697 void nfp_net_coalesce_write_cfg(struct nfp_net *nn)
2698 {
2699 	u8 i;
2700 	u32 factor;
2701 	u32 value;
2702 
2703 	/* Compute factor used to convert coalesce '_usecs' parameters to
2704 	 * ME timestamp ticks.  There are 16 ME clock cycles for each timestamp
2705 	 * count.
2706 	 */
2707 	factor = nn->tlv_caps.me_freq_mhz / 16;
2708 
2709 	/* copy RX interrupt coalesce parameters */
2710 	value = (nn->rx_coalesce_max_frames << 16) |
2711 		(factor * nn->rx_coalesce_usecs);
2712 	for (i = 0; i < nn->dp.num_rx_rings; i++)
2713 		nn_writel(nn, NFP_NET_CFG_RXR_IRQ_MOD(i), value);
2714 
2715 	/* copy TX interrupt coalesce parameters */
2716 	value = (nn->tx_coalesce_max_frames << 16) |
2717 		(factor * nn->tx_coalesce_usecs);
2718 	for (i = 0; i < nn->dp.num_tx_rings; i++)
2719 		nn_writel(nn, NFP_NET_CFG_TXR_IRQ_MOD(i), value);
2720 }
2721 
2722 /**
2723  * nfp_net_write_mac_addr() - Write mac address to the device control BAR
2724  * @nn:      NFP Net device to reconfigure
2725  * @addr:    MAC address to write
2726  *
2727  * Writes the MAC address from the netdev to the device control BAR.  Does not
2728  * perform the required reconfig.  We do a bit of byte swapping dance because
2729  * firmware is LE.
2730  */
2731 static void nfp_net_write_mac_addr(struct nfp_net *nn, const u8 *addr)
2732 {
2733 	nn_writel(nn, NFP_NET_CFG_MACADDR + 0, get_unaligned_be32(addr));
2734 	nn_writew(nn, NFP_NET_CFG_MACADDR + 6, get_unaligned_be16(addr + 4));
2735 }
2736 
2737 static void nfp_net_vec_clear_ring_data(struct nfp_net *nn, unsigned int idx)
2738 {
2739 	nn_writeq(nn, NFP_NET_CFG_RXR_ADDR(idx), 0);
2740 	nn_writeb(nn, NFP_NET_CFG_RXR_SZ(idx), 0);
2741 	nn_writeb(nn, NFP_NET_CFG_RXR_VEC(idx), 0);
2742 
2743 	nn_writeq(nn, NFP_NET_CFG_TXR_ADDR(idx), 0);
2744 	nn_writeb(nn, NFP_NET_CFG_TXR_SZ(idx), 0);
2745 	nn_writeb(nn, NFP_NET_CFG_TXR_VEC(idx), 0);
2746 }
2747 
2748 /**
2749  * nfp_net_clear_config_and_disable() - Clear control BAR and disable NFP
2750  * @nn:      NFP Net device to reconfigure
2751  *
2752  * Warning: must be fully idempotent.
2753  */
2754 static void nfp_net_clear_config_and_disable(struct nfp_net *nn)
2755 {
2756 	u32 new_ctrl, update;
2757 	unsigned int r;
2758 	int err;
2759 
2760 	new_ctrl = nn->dp.ctrl;
2761 	new_ctrl &= ~NFP_NET_CFG_CTRL_ENABLE;
2762 	update = NFP_NET_CFG_UPDATE_GEN;
2763 	update |= NFP_NET_CFG_UPDATE_MSIX;
2764 	update |= NFP_NET_CFG_UPDATE_RING;
2765 
2766 	if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG)
2767 		new_ctrl &= ~NFP_NET_CFG_CTRL_RINGCFG;
2768 
2769 	nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0);
2770 	nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0);
2771 
2772 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2773 	err = nfp_net_reconfig(nn, update);
2774 	if (err)
2775 		nn_err(nn, "Could not disable device: %d\n", err);
2776 
2777 	for (r = 0; r < nn->dp.num_rx_rings; r++)
2778 		nfp_net_rx_ring_reset(&nn->dp.rx_rings[r]);
2779 	for (r = 0; r < nn->dp.num_tx_rings; r++)
2780 		nfp_net_tx_ring_reset(&nn->dp, &nn->dp.tx_rings[r]);
2781 	for (r = 0; r < nn->dp.num_r_vecs; r++)
2782 		nfp_net_vec_clear_ring_data(nn, r);
2783 
2784 	nn->dp.ctrl = new_ctrl;
2785 }
2786 
2787 static void
2788 nfp_net_rx_ring_hw_cfg_write(struct nfp_net *nn,
2789 			     struct nfp_net_rx_ring *rx_ring, unsigned int idx)
2790 {
2791 	/* Write the DMA address, size and MSI-X info to the device */
2792 	nn_writeq(nn, NFP_NET_CFG_RXR_ADDR(idx), rx_ring->dma);
2793 	nn_writeb(nn, NFP_NET_CFG_RXR_SZ(idx), ilog2(rx_ring->cnt));
2794 	nn_writeb(nn, NFP_NET_CFG_RXR_VEC(idx), rx_ring->r_vec->irq_entry);
2795 }
2796 
2797 static void
2798 nfp_net_tx_ring_hw_cfg_write(struct nfp_net *nn,
2799 			     struct nfp_net_tx_ring *tx_ring, unsigned int idx)
2800 {
2801 	nn_writeq(nn, NFP_NET_CFG_TXR_ADDR(idx), tx_ring->dma);
2802 	nn_writeb(nn, NFP_NET_CFG_TXR_SZ(idx), ilog2(tx_ring->cnt));
2803 	nn_writeb(nn, NFP_NET_CFG_TXR_VEC(idx), tx_ring->r_vec->irq_entry);
2804 }
2805 
2806 /**
2807  * nfp_net_set_config_and_enable() - Write control BAR and enable NFP
2808  * @nn:      NFP Net device to reconfigure
2809  */
2810 static int nfp_net_set_config_and_enable(struct nfp_net *nn)
2811 {
2812 	u32 bufsz, new_ctrl, update = 0;
2813 	unsigned int r;
2814 	int err;
2815 
2816 	new_ctrl = nn->dp.ctrl;
2817 
2818 	if (nn->dp.ctrl & NFP_NET_CFG_CTRL_RSS_ANY) {
2819 		nfp_net_rss_write_key(nn);
2820 		nfp_net_rss_write_itbl(nn);
2821 		nn_writel(nn, NFP_NET_CFG_RSS_CTRL, nn->rss_cfg);
2822 		update |= NFP_NET_CFG_UPDATE_RSS;
2823 	}
2824 
2825 	if (nn->dp.ctrl & NFP_NET_CFG_CTRL_IRQMOD) {
2826 		nfp_net_coalesce_write_cfg(nn);
2827 		update |= NFP_NET_CFG_UPDATE_IRQMOD;
2828 	}
2829 
2830 	for (r = 0; r < nn->dp.num_tx_rings; r++)
2831 		nfp_net_tx_ring_hw_cfg_write(nn, &nn->dp.tx_rings[r], r);
2832 	for (r = 0; r < nn->dp.num_rx_rings; r++)
2833 		nfp_net_rx_ring_hw_cfg_write(nn, &nn->dp.rx_rings[r], r);
2834 
2835 	nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, nn->dp.num_tx_rings == 64 ?
2836 		  0xffffffffffffffffULL : ((u64)1 << nn->dp.num_tx_rings) - 1);
2837 
2838 	nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, nn->dp.num_rx_rings == 64 ?
2839 		  0xffffffffffffffffULL : ((u64)1 << nn->dp.num_rx_rings) - 1);
2840 
2841 	if (nn->dp.netdev)
2842 		nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr);
2843 
2844 	nn_writel(nn, NFP_NET_CFG_MTU, nn->dp.mtu);
2845 
2846 	bufsz = nn->dp.fl_bufsz - nn->dp.rx_dma_off - NFP_NET_RX_BUF_NON_DATA;
2847 	nn_writel(nn, NFP_NET_CFG_FLBUFSZ, bufsz);
2848 
2849 	/* Enable device */
2850 	new_ctrl |= NFP_NET_CFG_CTRL_ENABLE;
2851 	update |= NFP_NET_CFG_UPDATE_GEN;
2852 	update |= NFP_NET_CFG_UPDATE_MSIX;
2853 	update |= NFP_NET_CFG_UPDATE_RING;
2854 	if (nn->cap & NFP_NET_CFG_CTRL_RINGCFG)
2855 		new_ctrl |= NFP_NET_CFG_CTRL_RINGCFG;
2856 
2857 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
2858 	err = nfp_net_reconfig(nn, update);
2859 	if (err) {
2860 		nfp_net_clear_config_and_disable(nn);
2861 		return err;
2862 	}
2863 
2864 	nn->dp.ctrl = new_ctrl;
2865 
2866 	for (r = 0; r < nn->dp.num_rx_rings; r++)
2867 		nfp_net_rx_ring_fill_freelist(&nn->dp, &nn->dp.rx_rings[r]);
2868 
2869 	return 0;
2870 }
2871 
2872 /**
2873  * nfp_net_close_stack() - Quiesce the stack (part of close)
2874  * @nn:	     NFP Net device to reconfigure
2875  */
2876 static void nfp_net_close_stack(struct nfp_net *nn)
2877 {
2878 	unsigned int r;
2879 
2880 	disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2881 	netif_carrier_off(nn->dp.netdev);
2882 	nn->link_up = false;
2883 
2884 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
2885 		disable_irq(nn->r_vecs[r].irq_vector);
2886 		napi_disable(&nn->r_vecs[r].napi);
2887 	}
2888 
2889 	netif_tx_disable(nn->dp.netdev);
2890 }
2891 
2892 /**
2893  * nfp_net_close_free_all() - Free all runtime resources
2894  * @nn:      NFP Net device to reconfigure
2895  */
2896 static void nfp_net_close_free_all(struct nfp_net *nn)
2897 {
2898 	unsigned int r;
2899 
2900 	nfp_net_tx_rings_free(&nn->dp);
2901 	nfp_net_rx_rings_free(&nn->dp);
2902 
2903 	for (r = 0; r < nn->dp.num_r_vecs; r++)
2904 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
2905 
2906 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
2907 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX);
2908 }
2909 
2910 /**
2911  * nfp_net_netdev_close() - Called when the device is downed
2912  * @netdev:      netdev structure
2913  */
2914 static int nfp_net_netdev_close(struct net_device *netdev)
2915 {
2916 	struct nfp_net *nn = netdev_priv(netdev);
2917 
2918 	/* Step 1: Disable RX and TX rings from the Linux kernel perspective
2919 	 */
2920 	nfp_net_close_stack(nn);
2921 
2922 	/* Step 2: Tell NFP
2923 	 */
2924 	nfp_net_clear_config_and_disable(nn);
2925 	nfp_port_configure(netdev, false);
2926 
2927 	/* Step 3: Free resources
2928 	 */
2929 	nfp_net_close_free_all(nn);
2930 
2931 	nn_dbg(nn, "%s down", netdev->name);
2932 	return 0;
2933 }
2934 
2935 void nfp_ctrl_close(struct nfp_net *nn)
2936 {
2937 	int r;
2938 
2939 	rtnl_lock();
2940 
2941 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
2942 		disable_irq(nn->r_vecs[r].irq_vector);
2943 		tasklet_disable(&nn->r_vecs[r].tasklet);
2944 	}
2945 
2946 	nfp_net_clear_config_and_disable(nn);
2947 
2948 	nfp_net_close_free_all(nn);
2949 
2950 	rtnl_unlock();
2951 }
2952 
2953 /**
2954  * nfp_net_open_stack() - Start the device from stack's perspective
2955  * @nn:      NFP Net device to reconfigure
2956  */
2957 static void nfp_net_open_stack(struct nfp_net *nn)
2958 {
2959 	unsigned int r;
2960 
2961 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
2962 		napi_enable(&nn->r_vecs[r].napi);
2963 		enable_irq(nn->r_vecs[r].irq_vector);
2964 	}
2965 
2966 	netif_tx_wake_all_queues(nn->dp.netdev);
2967 
2968 	enable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2969 	nfp_net_read_link_status(nn);
2970 }
2971 
2972 static int nfp_net_open_alloc_all(struct nfp_net *nn)
2973 {
2974 	int err, r;
2975 
2976 	err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_EXN, "%s-exn",
2977 				      nn->exn_name, sizeof(nn->exn_name),
2978 				      NFP_NET_IRQ_EXN_IDX, nn->exn_handler);
2979 	if (err)
2980 		return err;
2981 	err = nfp_net_aux_irq_request(nn, NFP_NET_CFG_LSC, "%s-lsc",
2982 				      nn->lsc_name, sizeof(nn->lsc_name),
2983 				      NFP_NET_IRQ_LSC_IDX, nn->lsc_handler);
2984 	if (err)
2985 		goto err_free_exn;
2986 	disable_irq(nn->irq_entries[NFP_NET_IRQ_LSC_IDX].vector);
2987 
2988 	for (r = 0; r < nn->dp.num_r_vecs; r++) {
2989 		err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r);
2990 		if (err)
2991 			goto err_cleanup_vec_p;
2992 	}
2993 
2994 	err = nfp_net_rx_rings_prepare(nn, &nn->dp);
2995 	if (err)
2996 		goto err_cleanup_vec;
2997 
2998 	err = nfp_net_tx_rings_prepare(nn, &nn->dp);
2999 	if (err)
3000 		goto err_free_rx_rings;
3001 
3002 	for (r = 0; r < nn->max_r_vecs; r++)
3003 		nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r);
3004 
3005 	return 0;
3006 
3007 err_free_rx_rings:
3008 	nfp_net_rx_rings_free(&nn->dp);
3009 err_cleanup_vec:
3010 	r = nn->dp.num_r_vecs;
3011 err_cleanup_vec_p:
3012 	while (r--)
3013 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
3014 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_LSC, NFP_NET_IRQ_LSC_IDX);
3015 err_free_exn:
3016 	nfp_net_aux_irq_free(nn, NFP_NET_CFG_EXN, NFP_NET_IRQ_EXN_IDX);
3017 	return err;
3018 }
3019 
3020 static int nfp_net_netdev_open(struct net_device *netdev)
3021 {
3022 	struct nfp_net *nn = netdev_priv(netdev);
3023 	int err;
3024 
3025 	/* Step 1: Allocate resources for rings and the like
3026 	 * - Request interrupts
3027 	 * - Allocate RX and TX ring resources
3028 	 * - Setup initial RSS table
3029 	 */
3030 	err = nfp_net_open_alloc_all(nn);
3031 	if (err)
3032 		return err;
3033 
3034 	err = netif_set_real_num_tx_queues(netdev, nn->dp.num_stack_tx_rings);
3035 	if (err)
3036 		goto err_free_all;
3037 
3038 	err = netif_set_real_num_rx_queues(netdev, nn->dp.num_rx_rings);
3039 	if (err)
3040 		goto err_free_all;
3041 
3042 	/* Step 2: Configure the NFP
3043 	 * - Ifup the physical interface if it exists
3044 	 * - Enable rings from 0 to tx_rings/rx_rings - 1.
3045 	 * - Write MAC address (in case it changed)
3046 	 * - Set the MTU
3047 	 * - Set the Freelist buffer size
3048 	 * - Enable the FW
3049 	 */
3050 	err = nfp_port_configure(netdev, true);
3051 	if (err)
3052 		goto err_free_all;
3053 
3054 	err = nfp_net_set_config_and_enable(nn);
3055 	if (err)
3056 		goto err_port_disable;
3057 
3058 	/* Step 3: Enable for kernel
3059 	 * - put some freelist descriptors on each RX ring
3060 	 * - enable NAPI on each ring
3061 	 * - enable all TX queues
3062 	 * - set link state
3063 	 */
3064 	nfp_net_open_stack(nn);
3065 
3066 	return 0;
3067 
3068 err_port_disable:
3069 	nfp_port_configure(netdev, false);
3070 err_free_all:
3071 	nfp_net_close_free_all(nn);
3072 	return err;
3073 }
3074 
3075 int nfp_ctrl_open(struct nfp_net *nn)
3076 {
3077 	int err, r;
3078 
3079 	/* ring dumping depends on vNICs being opened/closed under rtnl */
3080 	rtnl_lock();
3081 
3082 	err = nfp_net_open_alloc_all(nn);
3083 	if (err)
3084 		goto err_unlock;
3085 
3086 	err = nfp_net_set_config_and_enable(nn);
3087 	if (err)
3088 		goto err_free_all;
3089 
3090 	for (r = 0; r < nn->dp.num_r_vecs; r++)
3091 		enable_irq(nn->r_vecs[r].irq_vector);
3092 
3093 	rtnl_unlock();
3094 
3095 	return 0;
3096 
3097 err_free_all:
3098 	nfp_net_close_free_all(nn);
3099 err_unlock:
3100 	rtnl_unlock();
3101 	return err;
3102 }
3103 
3104 static void nfp_net_set_rx_mode(struct net_device *netdev)
3105 {
3106 	struct nfp_net *nn = netdev_priv(netdev);
3107 	u32 new_ctrl;
3108 
3109 	new_ctrl = nn->dp.ctrl;
3110 
3111 	if (!netdev_mc_empty(netdev) || netdev->flags & IFF_ALLMULTI)
3112 		new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_L2MC;
3113 	else
3114 		new_ctrl &= ~NFP_NET_CFG_CTRL_L2MC;
3115 
3116 	if (netdev->flags & IFF_PROMISC) {
3117 		if (nn->cap & NFP_NET_CFG_CTRL_PROMISC)
3118 			new_ctrl |= NFP_NET_CFG_CTRL_PROMISC;
3119 		else
3120 			nn_warn(nn, "FW does not support promiscuous mode\n");
3121 	} else {
3122 		new_ctrl &= ~NFP_NET_CFG_CTRL_PROMISC;
3123 	}
3124 
3125 	if (new_ctrl == nn->dp.ctrl)
3126 		return;
3127 
3128 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
3129 	nfp_net_reconfig_post(nn, NFP_NET_CFG_UPDATE_GEN);
3130 
3131 	nn->dp.ctrl = new_ctrl;
3132 }
3133 
3134 static void nfp_net_rss_init_itbl(struct nfp_net *nn)
3135 {
3136 	int i;
3137 
3138 	for (i = 0; i < sizeof(nn->rss_itbl); i++)
3139 		nn->rss_itbl[i] =
3140 			ethtool_rxfh_indir_default(i, nn->dp.num_rx_rings);
3141 }
3142 
3143 static void nfp_net_dp_swap(struct nfp_net *nn, struct nfp_net_dp *dp)
3144 {
3145 	struct nfp_net_dp new_dp = *dp;
3146 
3147 	*dp = nn->dp;
3148 	nn->dp = new_dp;
3149 
3150 	nn->dp.netdev->mtu = new_dp.mtu;
3151 
3152 	if (!netif_is_rxfh_configured(nn->dp.netdev))
3153 		nfp_net_rss_init_itbl(nn);
3154 }
3155 
3156 static int nfp_net_dp_swap_enable(struct nfp_net *nn, struct nfp_net_dp *dp)
3157 {
3158 	unsigned int r;
3159 	int err;
3160 
3161 	nfp_net_dp_swap(nn, dp);
3162 
3163 	for (r = 0; r <	nn->max_r_vecs; r++)
3164 		nfp_net_vector_assign_rings(&nn->dp, &nn->r_vecs[r], r);
3165 
3166 	err = netif_set_real_num_rx_queues(nn->dp.netdev, nn->dp.num_rx_rings);
3167 	if (err)
3168 		return err;
3169 
3170 	if (nn->dp.netdev->real_num_tx_queues != nn->dp.num_stack_tx_rings) {
3171 		err = netif_set_real_num_tx_queues(nn->dp.netdev,
3172 						   nn->dp.num_stack_tx_rings);
3173 		if (err)
3174 			return err;
3175 	}
3176 
3177 	return nfp_net_set_config_and_enable(nn);
3178 }
3179 
3180 struct nfp_net_dp *nfp_net_clone_dp(struct nfp_net *nn)
3181 {
3182 	struct nfp_net_dp *new;
3183 
3184 	new = kmalloc(sizeof(*new), GFP_KERNEL);
3185 	if (!new)
3186 		return NULL;
3187 
3188 	*new = nn->dp;
3189 
3190 	/* Clear things which need to be recomputed */
3191 	new->fl_bufsz = 0;
3192 	new->tx_rings = NULL;
3193 	new->rx_rings = NULL;
3194 	new->num_r_vecs = 0;
3195 	new->num_stack_tx_rings = 0;
3196 
3197 	return new;
3198 }
3199 
3200 static int
3201 nfp_net_check_config(struct nfp_net *nn, struct nfp_net_dp *dp,
3202 		     struct netlink_ext_ack *extack)
3203 {
3204 	/* XDP-enabled tests */
3205 	if (!dp->xdp_prog)
3206 		return 0;
3207 	if (dp->fl_bufsz > PAGE_SIZE) {
3208 		NL_SET_ERR_MSG_MOD(extack, "MTU too large w/ XDP enabled");
3209 		return -EINVAL;
3210 	}
3211 	if (dp->num_tx_rings > nn->max_tx_rings) {
3212 		NL_SET_ERR_MSG_MOD(extack, "Insufficient number of TX rings w/ XDP enabled");
3213 		return -EINVAL;
3214 	}
3215 
3216 	return 0;
3217 }
3218 
3219 int nfp_net_ring_reconfig(struct nfp_net *nn, struct nfp_net_dp *dp,
3220 			  struct netlink_ext_ack *extack)
3221 {
3222 	int r, err;
3223 
3224 	dp->fl_bufsz = nfp_net_calc_fl_bufsz(dp);
3225 
3226 	dp->num_stack_tx_rings = dp->num_tx_rings;
3227 	if (dp->xdp_prog)
3228 		dp->num_stack_tx_rings -= dp->num_rx_rings;
3229 
3230 	dp->num_r_vecs = max(dp->num_rx_rings, dp->num_stack_tx_rings);
3231 
3232 	err = nfp_net_check_config(nn, dp, extack);
3233 	if (err)
3234 		goto exit_free_dp;
3235 
3236 	if (!netif_running(dp->netdev)) {
3237 		nfp_net_dp_swap(nn, dp);
3238 		err = 0;
3239 		goto exit_free_dp;
3240 	}
3241 
3242 	/* Prepare new rings */
3243 	for (r = nn->dp.num_r_vecs; r < dp->num_r_vecs; r++) {
3244 		err = nfp_net_prepare_vector(nn, &nn->r_vecs[r], r);
3245 		if (err) {
3246 			dp->num_r_vecs = r;
3247 			goto err_cleanup_vecs;
3248 		}
3249 	}
3250 
3251 	err = nfp_net_rx_rings_prepare(nn, dp);
3252 	if (err)
3253 		goto err_cleanup_vecs;
3254 
3255 	err = nfp_net_tx_rings_prepare(nn, dp);
3256 	if (err)
3257 		goto err_free_rx;
3258 
3259 	/* Stop device, swap in new rings, try to start the firmware */
3260 	nfp_net_close_stack(nn);
3261 	nfp_net_clear_config_and_disable(nn);
3262 
3263 	err = nfp_net_dp_swap_enable(nn, dp);
3264 	if (err) {
3265 		int err2;
3266 
3267 		nfp_net_clear_config_and_disable(nn);
3268 
3269 		/* Try with old configuration and old rings */
3270 		err2 = nfp_net_dp_swap_enable(nn, dp);
3271 		if (err2)
3272 			nn_err(nn, "Can't restore ring config - FW communication failed (%d,%d)\n",
3273 			       err, err2);
3274 	}
3275 	for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--)
3276 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
3277 
3278 	nfp_net_rx_rings_free(dp);
3279 	nfp_net_tx_rings_free(dp);
3280 
3281 	nfp_net_open_stack(nn);
3282 exit_free_dp:
3283 	kfree(dp);
3284 
3285 	return err;
3286 
3287 err_free_rx:
3288 	nfp_net_rx_rings_free(dp);
3289 err_cleanup_vecs:
3290 	for (r = dp->num_r_vecs - 1; r >= nn->dp.num_r_vecs; r--)
3291 		nfp_net_cleanup_vector(nn, &nn->r_vecs[r]);
3292 	kfree(dp);
3293 	return err;
3294 }
3295 
3296 static int nfp_net_change_mtu(struct net_device *netdev, int new_mtu)
3297 {
3298 	struct nfp_net *nn = netdev_priv(netdev);
3299 	struct nfp_net_dp *dp;
3300 	int err;
3301 
3302 	err = nfp_app_check_mtu(nn->app, netdev, new_mtu);
3303 	if (err)
3304 		return err;
3305 
3306 	dp = nfp_net_clone_dp(nn);
3307 	if (!dp)
3308 		return -ENOMEM;
3309 
3310 	dp->mtu = new_mtu;
3311 
3312 	return nfp_net_ring_reconfig(nn, dp, NULL);
3313 }
3314 
3315 static int
3316 nfp_net_vlan_rx_add_vid(struct net_device *netdev, __be16 proto, u16 vid)
3317 {
3318 	const u32 cmd = NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_ADD;
3319 	struct nfp_net *nn = netdev_priv(netdev);
3320 	int err;
3321 
3322 	/* Priority tagged packets with vlan id 0 are processed by the
3323 	 * NFP as untagged packets
3324 	 */
3325 	if (!vid)
3326 		return 0;
3327 
3328 	err = nfp_net_mbox_lock(nn, NFP_NET_CFG_VLAN_FILTER_SZ);
3329 	if (err)
3330 		return err;
3331 
3332 	nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_VID, vid);
3333 	nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_PROTO,
3334 		  ETH_P_8021Q);
3335 
3336 	return nfp_net_mbox_reconfig_and_unlock(nn, cmd);
3337 }
3338 
3339 static int
3340 nfp_net_vlan_rx_kill_vid(struct net_device *netdev, __be16 proto, u16 vid)
3341 {
3342 	const u32 cmd = NFP_NET_CFG_MBOX_CMD_CTAG_FILTER_KILL;
3343 	struct nfp_net *nn = netdev_priv(netdev);
3344 	int err;
3345 
3346 	/* Priority tagged packets with vlan id 0 are processed by the
3347 	 * NFP as untagged packets
3348 	 */
3349 	if (!vid)
3350 		return 0;
3351 
3352 	err = nfp_net_mbox_lock(nn, NFP_NET_CFG_VLAN_FILTER_SZ);
3353 	if (err)
3354 		return err;
3355 
3356 	nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_VID, vid);
3357 	nn_writew(nn, nn->tlv_caps.mbox_off + NFP_NET_CFG_VLAN_FILTER_PROTO,
3358 		  ETH_P_8021Q);
3359 
3360 	return nfp_net_mbox_reconfig_and_unlock(nn, cmd);
3361 }
3362 
3363 static void nfp_net_stat64(struct net_device *netdev,
3364 			   struct rtnl_link_stats64 *stats)
3365 {
3366 	struct nfp_net *nn = netdev_priv(netdev);
3367 	int r;
3368 
3369 	/* Collect software stats */
3370 	for (r = 0; r < nn->max_r_vecs; r++) {
3371 		struct nfp_net_r_vector *r_vec = &nn->r_vecs[r];
3372 		u64 data[3];
3373 		unsigned int start;
3374 
3375 		do {
3376 			start = u64_stats_fetch_begin(&r_vec->rx_sync);
3377 			data[0] = r_vec->rx_pkts;
3378 			data[1] = r_vec->rx_bytes;
3379 			data[2] = r_vec->rx_drops;
3380 		} while (u64_stats_fetch_retry(&r_vec->rx_sync, start));
3381 		stats->rx_packets += data[0];
3382 		stats->rx_bytes += data[1];
3383 		stats->rx_dropped += data[2];
3384 
3385 		do {
3386 			start = u64_stats_fetch_begin(&r_vec->tx_sync);
3387 			data[0] = r_vec->tx_pkts;
3388 			data[1] = r_vec->tx_bytes;
3389 			data[2] = r_vec->tx_errors;
3390 		} while (u64_stats_fetch_retry(&r_vec->tx_sync, start));
3391 		stats->tx_packets += data[0];
3392 		stats->tx_bytes += data[1];
3393 		stats->tx_errors += data[2];
3394 	}
3395 
3396 	/* Add in device stats */
3397 	stats->multicast += nn_readq(nn, NFP_NET_CFG_STATS_RX_MC_FRAMES);
3398 	stats->rx_dropped += nn_readq(nn, NFP_NET_CFG_STATS_RX_DISCARDS);
3399 	stats->rx_errors += nn_readq(nn, NFP_NET_CFG_STATS_RX_ERRORS);
3400 
3401 	stats->tx_dropped += nn_readq(nn, NFP_NET_CFG_STATS_TX_DISCARDS);
3402 	stats->tx_errors += nn_readq(nn, NFP_NET_CFG_STATS_TX_ERRORS);
3403 }
3404 
3405 static int nfp_net_set_features(struct net_device *netdev,
3406 				netdev_features_t features)
3407 {
3408 	netdev_features_t changed = netdev->features ^ features;
3409 	struct nfp_net *nn = netdev_priv(netdev);
3410 	u32 new_ctrl;
3411 	int err;
3412 
3413 	/* Assume this is not called with features we have not advertised */
3414 
3415 	new_ctrl = nn->dp.ctrl;
3416 
3417 	if (changed & NETIF_F_RXCSUM) {
3418 		if (features & NETIF_F_RXCSUM)
3419 			new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY;
3420 		else
3421 			new_ctrl &= ~NFP_NET_CFG_CTRL_RXCSUM_ANY;
3422 	}
3423 
3424 	if (changed & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM)) {
3425 		if (features & (NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM))
3426 			new_ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
3427 		else
3428 			new_ctrl &= ~NFP_NET_CFG_CTRL_TXCSUM;
3429 	}
3430 
3431 	if (changed & (NETIF_F_TSO | NETIF_F_TSO6)) {
3432 		if (features & (NETIF_F_TSO | NETIF_F_TSO6))
3433 			new_ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?:
3434 					      NFP_NET_CFG_CTRL_LSO;
3435 		else
3436 			new_ctrl &= ~NFP_NET_CFG_CTRL_LSO_ANY;
3437 	}
3438 
3439 	if (changed & NETIF_F_HW_VLAN_CTAG_RX) {
3440 		if (features & NETIF_F_HW_VLAN_CTAG_RX)
3441 			new_ctrl |= NFP_NET_CFG_CTRL_RXVLAN;
3442 		else
3443 			new_ctrl &= ~NFP_NET_CFG_CTRL_RXVLAN;
3444 	}
3445 
3446 	if (changed & NETIF_F_HW_VLAN_CTAG_TX) {
3447 		if (features & NETIF_F_HW_VLAN_CTAG_TX)
3448 			new_ctrl |= NFP_NET_CFG_CTRL_TXVLAN;
3449 		else
3450 			new_ctrl &= ~NFP_NET_CFG_CTRL_TXVLAN;
3451 	}
3452 
3453 	if (changed & NETIF_F_HW_VLAN_CTAG_FILTER) {
3454 		if (features & NETIF_F_HW_VLAN_CTAG_FILTER)
3455 			new_ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER;
3456 		else
3457 			new_ctrl &= ~NFP_NET_CFG_CTRL_CTAG_FILTER;
3458 	}
3459 
3460 	if (changed & NETIF_F_SG) {
3461 		if (features & NETIF_F_SG)
3462 			new_ctrl |= NFP_NET_CFG_CTRL_GATHER;
3463 		else
3464 			new_ctrl &= ~NFP_NET_CFG_CTRL_GATHER;
3465 	}
3466 
3467 	err = nfp_port_set_features(netdev, features);
3468 	if (err)
3469 		return err;
3470 
3471 	nn_dbg(nn, "Feature change 0x%llx -> 0x%llx (changed=0x%llx)\n",
3472 	       netdev->features, features, changed);
3473 
3474 	if (new_ctrl == nn->dp.ctrl)
3475 		return 0;
3476 
3477 	nn_dbg(nn, "NIC ctrl: 0x%x -> 0x%x\n", nn->dp.ctrl, new_ctrl);
3478 	nn_writel(nn, NFP_NET_CFG_CTRL, new_ctrl);
3479 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_GEN);
3480 	if (err)
3481 		return err;
3482 
3483 	nn->dp.ctrl = new_ctrl;
3484 
3485 	return 0;
3486 }
3487 
3488 static netdev_features_t
3489 nfp_net_features_check(struct sk_buff *skb, struct net_device *dev,
3490 		       netdev_features_t features)
3491 {
3492 	u8 l4_hdr;
3493 
3494 	/* We can't do TSO over double tagged packets (802.1AD) */
3495 	features &= vlan_features_check(skb, features);
3496 
3497 	if (!skb->encapsulation)
3498 		return features;
3499 
3500 	/* Ensure that inner L4 header offset fits into TX descriptor field */
3501 	if (skb_is_gso(skb)) {
3502 		u32 hdrlen;
3503 
3504 		hdrlen = skb_inner_transport_header(skb) - skb->data +
3505 			inner_tcp_hdrlen(skb);
3506 
3507 		/* Assume worst case scenario of having longest possible
3508 		 * metadata prepend - 8B
3509 		 */
3510 		if (unlikely(hdrlen > NFP_NET_LSO_MAX_HDR_SZ - 8))
3511 			features &= ~NETIF_F_GSO_MASK;
3512 	}
3513 
3514 	/* VXLAN/GRE check */
3515 	switch (vlan_get_protocol(skb)) {
3516 	case htons(ETH_P_IP):
3517 		l4_hdr = ip_hdr(skb)->protocol;
3518 		break;
3519 	case htons(ETH_P_IPV6):
3520 		l4_hdr = ipv6_hdr(skb)->nexthdr;
3521 		break;
3522 	default:
3523 		return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3524 	}
3525 
3526 	if (skb->inner_protocol_type != ENCAP_TYPE_ETHER ||
3527 	    skb->inner_protocol != htons(ETH_P_TEB) ||
3528 	    (l4_hdr != IPPROTO_UDP && l4_hdr != IPPROTO_GRE) ||
3529 	    (l4_hdr == IPPROTO_UDP &&
3530 	     (skb_inner_mac_header(skb) - skb_transport_header(skb) !=
3531 	      sizeof(struct udphdr) + sizeof(struct vxlanhdr))))
3532 		return features & ~(NETIF_F_CSUM_MASK | NETIF_F_GSO_MASK);
3533 
3534 	return features;
3535 }
3536 
3537 static int
3538 nfp_net_get_phys_port_name(struct net_device *netdev, char *name, size_t len)
3539 {
3540 	struct nfp_net *nn = netdev_priv(netdev);
3541 	int n;
3542 
3543 	/* If port is defined, devlink_port is registered and devlink core
3544 	 * is taking care of name formatting.
3545 	 */
3546 	if (nn->port)
3547 		return -EOPNOTSUPP;
3548 
3549 	if (nn->dp.is_vf || nn->vnic_no_name)
3550 		return -EOPNOTSUPP;
3551 
3552 	n = snprintf(name, len, "n%d", nn->id);
3553 	if (n >= len)
3554 		return -EINVAL;
3555 
3556 	return 0;
3557 }
3558 
3559 static int nfp_net_xdp_setup_drv(struct nfp_net *nn, struct netdev_bpf *bpf)
3560 {
3561 	struct bpf_prog *prog = bpf->prog;
3562 	struct nfp_net_dp *dp;
3563 	int err;
3564 
3565 	if (!prog == !nn->dp.xdp_prog) {
3566 		WRITE_ONCE(nn->dp.xdp_prog, prog);
3567 		xdp_attachment_setup(&nn->xdp, bpf);
3568 		return 0;
3569 	}
3570 
3571 	dp = nfp_net_clone_dp(nn);
3572 	if (!dp)
3573 		return -ENOMEM;
3574 
3575 	dp->xdp_prog = prog;
3576 	dp->num_tx_rings += prog ? nn->dp.num_rx_rings : -nn->dp.num_rx_rings;
3577 	dp->rx_dma_dir = prog ? DMA_BIDIRECTIONAL : DMA_FROM_DEVICE;
3578 	dp->rx_dma_off = prog ? XDP_PACKET_HEADROOM - nn->dp.rx_offset : 0;
3579 
3580 	/* We need RX reconfig to remap the buffers (BIDIR vs FROM_DEV) */
3581 	err = nfp_net_ring_reconfig(nn, dp, bpf->extack);
3582 	if (err)
3583 		return err;
3584 
3585 	xdp_attachment_setup(&nn->xdp, bpf);
3586 	return 0;
3587 }
3588 
3589 static int nfp_net_xdp_setup_hw(struct nfp_net *nn, struct netdev_bpf *bpf)
3590 {
3591 	int err;
3592 
3593 	err = nfp_app_xdp_offload(nn->app, nn, bpf->prog, bpf->extack);
3594 	if (err)
3595 		return err;
3596 
3597 	xdp_attachment_setup(&nn->xdp_hw, bpf);
3598 	return 0;
3599 }
3600 
3601 static int nfp_net_xdp(struct net_device *netdev, struct netdev_bpf *xdp)
3602 {
3603 	struct nfp_net *nn = netdev_priv(netdev);
3604 
3605 	switch (xdp->command) {
3606 	case XDP_SETUP_PROG:
3607 		return nfp_net_xdp_setup_drv(nn, xdp);
3608 	case XDP_SETUP_PROG_HW:
3609 		return nfp_net_xdp_setup_hw(nn, xdp);
3610 	default:
3611 		return nfp_app_bpf(nn->app, nn, xdp);
3612 	}
3613 }
3614 
3615 static int nfp_net_set_mac_address(struct net_device *netdev, void *addr)
3616 {
3617 	struct nfp_net *nn = netdev_priv(netdev);
3618 	struct sockaddr *saddr = addr;
3619 	int err;
3620 
3621 	err = eth_prepare_mac_addr_change(netdev, addr);
3622 	if (err)
3623 		return err;
3624 
3625 	nfp_net_write_mac_addr(nn, saddr->sa_data);
3626 
3627 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_MACADDR);
3628 	if (err)
3629 		return err;
3630 
3631 	eth_commit_mac_addr_change(netdev, addr);
3632 
3633 	return 0;
3634 }
3635 
3636 const struct net_device_ops nfp_net_netdev_ops = {
3637 	.ndo_init		= nfp_app_ndo_init,
3638 	.ndo_uninit		= nfp_app_ndo_uninit,
3639 	.ndo_open		= nfp_net_netdev_open,
3640 	.ndo_stop		= nfp_net_netdev_close,
3641 	.ndo_start_xmit		= nfp_net_tx,
3642 	.ndo_get_stats64	= nfp_net_stat64,
3643 	.ndo_vlan_rx_add_vid	= nfp_net_vlan_rx_add_vid,
3644 	.ndo_vlan_rx_kill_vid	= nfp_net_vlan_rx_kill_vid,
3645 	.ndo_set_vf_mac         = nfp_app_set_vf_mac,
3646 	.ndo_set_vf_vlan        = nfp_app_set_vf_vlan,
3647 	.ndo_set_vf_spoofchk    = nfp_app_set_vf_spoofchk,
3648 	.ndo_set_vf_trust	= nfp_app_set_vf_trust,
3649 	.ndo_get_vf_config	= nfp_app_get_vf_config,
3650 	.ndo_set_vf_link_state  = nfp_app_set_vf_link_state,
3651 	.ndo_setup_tc		= nfp_port_setup_tc,
3652 	.ndo_tx_timeout		= nfp_net_tx_timeout,
3653 	.ndo_set_rx_mode	= nfp_net_set_rx_mode,
3654 	.ndo_change_mtu		= nfp_net_change_mtu,
3655 	.ndo_set_mac_address	= nfp_net_set_mac_address,
3656 	.ndo_set_features	= nfp_net_set_features,
3657 	.ndo_features_check	= nfp_net_features_check,
3658 	.ndo_get_phys_port_name	= nfp_net_get_phys_port_name,
3659 	.ndo_bpf		= nfp_net_xdp,
3660 	.ndo_get_devlink_port	= nfp_devlink_get_devlink_port,
3661 };
3662 
3663 static int nfp_udp_tunnel_sync(struct net_device *netdev, unsigned int table)
3664 {
3665 	struct nfp_net *nn = netdev_priv(netdev);
3666 	int i;
3667 
3668 	BUILD_BUG_ON(NFP_NET_N_VXLAN_PORTS & 1);
3669 	for (i = 0; i < NFP_NET_N_VXLAN_PORTS; i += 2) {
3670 		struct udp_tunnel_info ti0, ti1;
3671 
3672 		udp_tunnel_nic_get_port(netdev, table, i, &ti0);
3673 		udp_tunnel_nic_get_port(netdev, table, i + 1, &ti1);
3674 
3675 		nn_writel(nn, NFP_NET_CFG_VXLAN_PORT + i * sizeof(ti0.port),
3676 			  be16_to_cpu(ti1.port) << 16 | be16_to_cpu(ti0.port));
3677 	}
3678 
3679 	return nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_VXLAN);
3680 }
3681 
3682 static const struct udp_tunnel_nic_info nfp_udp_tunnels = {
3683 	.sync_table     = nfp_udp_tunnel_sync,
3684 	.flags          = UDP_TUNNEL_NIC_INFO_MAY_SLEEP |
3685 			  UDP_TUNNEL_NIC_INFO_OPEN_ONLY,
3686 	.tables         = {
3687 		{
3688 			.n_entries      = NFP_NET_N_VXLAN_PORTS,
3689 			.tunnel_types   = UDP_TUNNEL_TYPE_VXLAN,
3690 		},
3691 	},
3692 };
3693 
3694 /**
3695  * nfp_net_info() - Print general info about the NIC
3696  * @nn:      NFP Net device to reconfigure
3697  */
3698 void nfp_net_info(struct nfp_net *nn)
3699 {
3700 	nn_info(nn, "Netronome NFP-6xxx %sNetdev: TxQs=%d/%d RxQs=%d/%d\n",
3701 		nn->dp.is_vf ? "VF " : "",
3702 		nn->dp.num_tx_rings, nn->max_tx_rings,
3703 		nn->dp.num_rx_rings, nn->max_rx_rings);
3704 	nn_info(nn, "VER: %d.%d.%d.%d, Maximum supported MTU: %d\n",
3705 		nn->fw_ver.resv, nn->fw_ver.class,
3706 		nn->fw_ver.major, nn->fw_ver.minor,
3707 		nn->max_mtu);
3708 	nn_info(nn, "CAP: %#x %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s\n",
3709 		nn->cap,
3710 		nn->cap & NFP_NET_CFG_CTRL_PROMISC  ? "PROMISC "  : "",
3711 		nn->cap & NFP_NET_CFG_CTRL_L2BC     ? "L2BCFILT " : "",
3712 		nn->cap & NFP_NET_CFG_CTRL_L2MC     ? "L2MCFILT " : "",
3713 		nn->cap & NFP_NET_CFG_CTRL_RXCSUM   ? "RXCSUM "   : "",
3714 		nn->cap & NFP_NET_CFG_CTRL_TXCSUM   ? "TXCSUM "   : "",
3715 		nn->cap & NFP_NET_CFG_CTRL_RXVLAN   ? "RXVLAN "   : "",
3716 		nn->cap & NFP_NET_CFG_CTRL_TXVLAN   ? "TXVLAN "   : "",
3717 		nn->cap & NFP_NET_CFG_CTRL_SCATTER  ? "SCATTER "  : "",
3718 		nn->cap & NFP_NET_CFG_CTRL_GATHER   ? "GATHER "   : "",
3719 		nn->cap & NFP_NET_CFG_CTRL_LSO      ? "TSO1 "     : "",
3720 		nn->cap & NFP_NET_CFG_CTRL_LSO2     ? "TSO2 "     : "",
3721 		nn->cap & NFP_NET_CFG_CTRL_RSS      ? "RSS1 "     : "",
3722 		nn->cap & NFP_NET_CFG_CTRL_RSS2     ? "RSS2 "     : "",
3723 		nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER ? "CTAG_FILTER " : "",
3724 		nn->cap & NFP_NET_CFG_CTRL_MSIXAUTO ? "AUTOMASK " : "",
3725 		nn->cap & NFP_NET_CFG_CTRL_IRQMOD   ? "IRQMOD "   : "",
3726 		nn->cap & NFP_NET_CFG_CTRL_VXLAN    ? "VXLAN "    : "",
3727 		nn->cap & NFP_NET_CFG_CTRL_NVGRE    ? "NVGRE "	  : "",
3728 		nn->cap & NFP_NET_CFG_CTRL_CSUM_COMPLETE ?
3729 						      "RXCSUM_COMPLETE " : "",
3730 		nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR ? "LIVE_ADDR " : "",
3731 		nfp_app_extra_cap(nn->app, nn));
3732 }
3733 
3734 /**
3735  * nfp_net_alloc() - Allocate netdev and related structure
3736  * @pdev:         PCI device
3737  * @ctrl_bar:     PCI IOMEM with vNIC config memory
3738  * @needs_netdev: Whether to allocate a netdev for this vNIC
3739  * @max_tx_rings: Maximum number of TX rings supported by device
3740  * @max_rx_rings: Maximum number of RX rings supported by device
3741  *
3742  * This function allocates a netdev device and fills in the initial
3743  * part of the @struct nfp_net structure.  In case of control device
3744  * nfp_net structure is allocated without the netdev.
3745  *
3746  * Return: NFP Net device structure, or ERR_PTR on error.
3747  */
3748 struct nfp_net *
3749 nfp_net_alloc(struct pci_dev *pdev, void __iomem *ctrl_bar, bool needs_netdev,
3750 	      unsigned int max_tx_rings, unsigned int max_rx_rings)
3751 {
3752 	struct nfp_net *nn;
3753 	int err;
3754 
3755 	if (needs_netdev) {
3756 		struct net_device *netdev;
3757 
3758 		netdev = alloc_etherdev_mqs(sizeof(struct nfp_net),
3759 					    max_tx_rings, max_rx_rings);
3760 		if (!netdev)
3761 			return ERR_PTR(-ENOMEM);
3762 
3763 		SET_NETDEV_DEV(netdev, &pdev->dev);
3764 		nn = netdev_priv(netdev);
3765 		nn->dp.netdev = netdev;
3766 	} else {
3767 		nn = vzalloc(sizeof(*nn));
3768 		if (!nn)
3769 			return ERR_PTR(-ENOMEM);
3770 	}
3771 
3772 	nn->dp.dev = &pdev->dev;
3773 	nn->dp.ctrl_bar = ctrl_bar;
3774 	nn->pdev = pdev;
3775 
3776 	nn->max_tx_rings = max_tx_rings;
3777 	nn->max_rx_rings = max_rx_rings;
3778 
3779 	nn->dp.num_tx_rings = min_t(unsigned int,
3780 				    max_tx_rings, num_online_cpus());
3781 	nn->dp.num_rx_rings = min_t(unsigned int, max_rx_rings,
3782 				 netif_get_num_default_rss_queues());
3783 
3784 	nn->dp.num_r_vecs = max(nn->dp.num_tx_rings, nn->dp.num_rx_rings);
3785 	nn->dp.num_r_vecs = min_t(unsigned int,
3786 				  nn->dp.num_r_vecs, num_online_cpus());
3787 
3788 	nn->dp.txd_cnt = NFP_NET_TX_DESCS_DEFAULT;
3789 	nn->dp.rxd_cnt = NFP_NET_RX_DESCS_DEFAULT;
3790 
3791 	sema_init(&nn->bar_lock, 1);
3792 
3793 	spin_lock_init(&nn->reconfig_lock);
3794 	spin_lock_init(&nn->link_status_lock);
3795 
3796 	timer_setup(&nn->reconfig_timer, nfp_net_reconfig_timer, 0);
3797 
3798 	err = nfp_net_tlv_caps_parse(&nn->pdev->dev, nn->dp.ctrl_bar,
3799 				     &nn->tlv_caps);
3800 	if (err)
3801 		goto err_free_nn;
3802 
3803 	err = nfp_ccm_mbox_alloc(nn);
3804 	if (err)
3805 		goto err_free_nn;
3806 
3807 	return nn;
3808 
3809 err_free_nn:
3810 	if (nn->dp.netdev)
3811 		free_netdev(nn->dp.netdev);
3812 	else
3813 		vfree(nn);
3814 	return ERR_PTR(err);
3815 }
3816 
3817 /**
3818  * nfp_net_free() - Undo what @nfp_net_alloc() did
3819  * @nn:      NFP Net device to reconfigure
3820  */
3821 void nfp_net_free(struct nfp_net *nn)
3822 {
3823 	WARN_ON(timer_pending(&nn->reconfig_timer) || nn->reconfig_posted);
3824 	nfp_ccm_mbox_free(nn);
3825 
3826 	if (nn->dp.netdev)
3827 		free_netdev(nn->dp.netdev);
3828 	else
3829 		vfree(nn);
3830 }
3831 
3832 /**
3833  * nfp_net_rss_key_sz() - Get current size of the RSS key
3834  * @nn:		NFP Net device instance
3835  *
3836  * Return: size of the RSS key for currently selected hash function.
3837  */
3838 unsigned int nfp_net_rss_key_sz(struct nfp_net *nn)
3839 {
3840 	switch (nn->rss_hfunc) {
3841 	case ETH_RSS_HASH_TOP:
3842 		return NFP_NET_CFG_RSS_KEY_SZ;
3843 	case ETH_RSS_HASH_XOR:
3844 		return 0;
3845 	case ETH_RSS_HASH_CRC32:
3846 		return 4;
3847 	}
3848 
3849 	nn_warn(nn, "Unknown hash function: %u\n", nn->rss_hfunc);
3850 	return 0;
3851 }
3852 
3853 /**
3854  * nfp_net_rss_init() - Set the initial RSS parameters
3855  * @nn:	     NFP Net device to reconfigure
3856  */
3857 static void nfp_net_rss_init(struct nfp_net *nn)
3858 {
3859 	unsigned long func_bit, rss_cap_hfunc;
3860 	u32 reg;
3861 
3862 	/* Read the RSS function capability and select first supported func */
3863 	reg = nn_readl(nn, NFP_NET_CFG_RSS_CAP);
3864 	rss_cap_hfunc =	FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC, reg);
3865 	if (!rss_cap_hfunc)
3866 		rss_cap_hfunc =	FIELD_GET(NFP_NET_CFG_RSS_CAP_HFUNC,
3867 					  NFP_NET_CFG_RSS_TOEPLITZ);
3868 
3869 	func_bit = find_first_bit(&rss_cap_hfunc, NFP_NET_CFG_RSS_HFUNCS);
3870 	if (func_bit == NFP_NET_CFG_RSS_HFUNCS) {
3871 		dev_warn(nn->dp.dev,
3872 			 "Bad RSS config, defaulting to Toeplitz hash\n");
3873 		func_bit = ETH_RSS_HASH_TOP_BIT;
3874 	}
3875 	nn->rss_hfunc = 1 << func_bit;
3876 
3877 	netdev_rss_key_fill(nn->rss_key, nfp_net_rss_key_sz(nn));
3878 
3879 	nfp_net_rss_init_itbl(nn);
3880 
3881 	/* Enable IPv4/IPv6 TCP by default */
3882 	nn->rss_cfg = NFP_NET_CFG_RSS_IPV4_TCP |
3883 		      NFP_NET_CFG_RSS_IPV6_TCP |
3884 		      FIELD_PREP(NFP_NET_CFG_RSS_HFUNC, nn->rss_hfunc) |
3885 		      NFP_NET_CFG_RSS_MASK;
3886 }
3887 
3888 /**
3889  * nfp_net_irqmod_init() - Set the initial IRQ moderation parameters
3890  * @nn:	     NFP Net device to reconfigure
3891  */
3892 static void nfp_net_irqmod_init(struct nfp_net *nn)
3893 {
3894 	nn->rx_coalesce_usecs      = 50;
3895 	nn->rx_coalesce_max_frames = 64;
3896 	nn->tx_coalesce_usecs      = 50;
3897 	nn->tx_coalesce_max_frames = 64;
3898 }
3899 
3900 static void nfp_net_netdev_init(struct nfp_net *nn)
3901 {
3902 	struct net_device *netdev = nn->dp.netdev;
3903 
3904 	nfp_net_write_mac_addr(nn, nn->dp.netdev->dev_addr);
3905 
3906 	netdev->mtu = nn->dp.mtu;
3907 
3908 	/* Advertise/enable offloads based on capabilities
3909 	 *
3910 	 * Note: netdev->features show the currently enabled features
3911 	 * and netdev->hw_features advertises which features are
3912 	 * supported.  By default we enable most features.
3913 	 */
3914 	if (nn->cap & NFP_NET_CFG_CTRL_LIVE_ADDR)
3915 		netdev->priv_flags |= IFF_LIVE_ADDR_CHANGE;
3916 
3917 	netdev->hw_features = NETIF_F_HIGHDMA;
3918 	if (nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY) {
3919 		netdev->hw_features |= NETIF_F_RXCSUM;
3920 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RXCSUM_ANY;
3921 	}
3922 	if (nn->cap & NFP_NET_CFG_CTRL_TXCSUM) {
3923 		netdev->hw_features |= NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM;
3924 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXCSUM;
3925 	}
3926 	if (nn->cap & NFP_NET_CFG_CTRL_GATHER) {
3927 		netdev->hw_features |= NETIF_F_SG;
3928 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_GATHER;
3929 	}
3930 	if ((nn->cap & NFP_NET_CFG_CTRL_LSO && nn->fw_ver.major > 2) ||
3931 	    nn->cap & NFP_NET_CFG_CTRL_LSO2) {
3932 		netdev->hw_features |= NETIF_F_TSO | NETIF_F_TSO6;
3933 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_LSO2 ?:
3934 					 NFP_NET_CFG_CTRL_LSO;
3935 	}
3936 	if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY)
3937 		netdev->hw_features |= NETIF_F_RXHASH;
3938 	if (nn->cap & NFP_NET_CFG_CTRL_VXLAN) {
3939 		if (nn->cap & NFP_NET_CFG_CTRL_LSO)
3940 			netdev->hw_features |= NETIF_F_GSO_UDP_TUNNEL;
3941 		netdev->udp_tunnel_nic_info = &nfp_udp_tunnels;
3942 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_VXLAN;
3943 	}
3944 	if (nn->cap & NFP_NET_CFG_CTRL_NVGRE) {
3945 		if (nn->cap & NFP_NET_CFG_CTRL_LSO)
3946 			netdev->hw_features |= NETIF_F_GSO_GRE;
3947 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_NVGRE;
3948 	}
3949 	if (nn->cap & (NFP_NET_CFG_CTRL_VXLAN | NFP_NET_CFG_CTRL_NVGRE))
3950 		netdev->hw_enc_features = netdev->hw_features;
3951 
3952 	netdev->vlan_features = netdev->hw_features;
3953 
3954 	if (nn->cap & NFP_NET_CFG_CTRL_RXVLAN) {
3955 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_RX;
3956 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_RXVLAN;
3957 	}
3958 	if (nn->cap & NFP_NET_CFG_CTRL_TXVLAN) {
3959 		if (nn->cap & NFP_NET_CFG_CTRL_LSO2) {
3960 			nn_warn(nn, "Device advertises both TSO2 and TXVLAN. Refusing to enable TXVLAN.\n");
3961 		} else {
3962 			netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_TX;
3963 			nn->dp.ctrl |= NFP_NET_CFG_CTRL_TXVLAN;
3964 		}
3965 	}
3966 	if (nn->cap & NFP_NET_CFG_CTRL_CTAG_FILTER) {
3967 		netdev->hw_features |= NETIF_F_HW_VLAN_CTAG_FILTER;
3968 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_CTAG_FILTER;
3969 	}
3970 
3971 	netdev->features = netdev->hw_features;
3972 
3973 	if (nfp_app_has_tc(nn->app) && nn->port)
3974 		netdev->hw_features |= NETIF_F_HW_TC;
3975 
3976 	/* Advertise but disable TSO by default. */
3977 	netdev->features &= ~(NETIF_F_TSO | NETIF_F_TSO6);
3978 	nn->dp.ctrl &= ~NFP_NET_CFG_CTRL_LSO_ANY;
3979 
3980 	/* Finalise the netdev setup */
3981 	netdev->netdev_ops = &nfp_net_netdev_ops;
3982 	netdev->watchdog_timeo = msecs_to_jiffies(5 * 1000);
3983 
3984 	/* MTU range: 68 - hw-specific max */
3985 	netdev->min_mtu = ETH_MIN_MTU;
3986 	netdev->max_mtu = nn->max_mtu;
3987 
3988 	netdev->gso_max_segs = NFP_NET_LSO_MAX_SEGS;
3989 
3990 	netif_carrier_off(netdev);
3991 
3992 	nfp_net_set_ethtool_ops(netdev);
3993 }
3994 
3995 static int nfp_net_read_caps(struct nfp_net *nn)
3996 {
3997 	/* Get some of the read-only fields from the BAR */
3998 	nn->cap = nn_readl(nn, NFP_NET_CFG_CAP);
3999 	nn->max_mtu = nn_readl(nn, NFP_NET_CFG_MAX_MTU);
4000 
4001 	/* ABI 4.x and ctrl vNIC always use chained metadata, in other cases
4002 	 * we allow use of non-chained metadata if RSS(v1) is the only
4003 	 * advertised capability requiring metadata.
4004 	 */
4005 	nn->dp.chained_metadata_format = nn->fw_ver.major == 4 ||
4006 					 !nn->dp.netdev ||
4007 					 !(nn->cap & NFP_NET_CFG_CTRL_RSS) ||
4008 					 nn->cap & NFP_NET_CFG_CTRL_CHAIN_META;
4009 	/* RSS(v1) uses non-chained metadata format, except in ABI 4.x where
4010 	 * it has the same meaning as RSSv2.
4011 	 */
4012 	if (nn->dp.chained_metadata_format && nn->fw_ver.major != 4)
4013 		nn->cap &= ~NFP_NET_CFG_CTRL_RSS;
4014 
4015 	/* Determine RX packet/metadata boundary offset */
4016 	if (nn->fw_ver.major >= 2) {
4017 		u32 reg;
4018 
4019 		reg = nn_readl(nn, NFP_NET_CFG_RX_OFFSET);
4020 		if (reg > NFP_NET_MAX_PREPEND) {
4021 			nn_err(nn, "Invalid rx offset: %d\n", reg);
4022 			return -EINVAL;
4023 		}
4024 		nn->dp.rx_offset = reg;
4025 	} else {
4026 		nn->dp.rx_offset = NFP_NET_RX_OFFSET;
4027 	}
4028 
4029 	/* For control vNICs mask out the capabilities app doesn't want. */
4030 	if (!nn->dp.netdev)
4031 		nn->cap &= nn->app->type->ctrl_cap_mask;
4032 
4033 	return 0;
4034 }
4035 
4036 /**
4037  * nfp_net_init() - Initialise/finalise the nfp_net structure
4038  * @nn:		NFP Net device structure
4039  *
4040  * Return: 0 on success or negative errno on error.
4041  */
4042 int nfp_net_init(struct nfp_net *nn)
4043 {
4044 	int err;
4045 
4046 	nn->dp.rx_dma_dir = DMA_FROM_DEVICE;
4047 
4048 	err = nfp_net_read_caps(nn);
4049 	if (err)
4050 		return err;
4051 
4052 	/* Set default MTU and Freelist buffer size */
4053 	if (!nfp_net_is_data_vnic(nn) && nn->app->ctrl_mtu) {
4054 		nn->dp.mtu = min(nn->app->ctrl_mtu, nn->max_mtu);
4055 	} else if (nn->max_mtu < NFP_NET_DEFAULT_MTU) {
4056 		nn->dp.mtu = nn->max_mtu;
4057 	} else {
4058 		nn->dp.mtu = NFP_NET_DEFAULT_MTU;
4059 	}
4060 	nn->dp.fl_bufsz = nfp_net_calc_fl_bufsz(&nn->dp);
4061 
4062 	if (nfp_app_ctrl_uses_data_vnics(nn->app))
4063 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_CMSG_DATA;
4064 
4065 	if (nn->cap & NFP_NET_CFG_CTRL_RSS_ANY) {
4066 		nfp_net_rss_init(nn);
4067 		nn->dp.ctrl |= nn->cap & NFP_NET_CFG_CTRL_RSS2 ?:
4068 					 NFP_NET_CFG_CTRL_RSS;
4069 	}
4070 
4071 	/* Allow L2 Broadcast and Multicast through by default, if supported */
4072 	if (nn->cap & NFP_NET_CFG_CTRL_L2BC)
4073 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_L2BC;
4074 
4075 	/* Allow IRQ moderation, if supported */
4076 	if (nn->cap & NFP_NET_CFG_CTRL_IRQMOD) {
4077 		nfp_net_irqmod_init(nn);
4078 		nn->dp.ctrl |= NFP_NET_CFG_CTRL_IRQMOD;
4079 	}
4080 
4081 	/* Stash the re-configuration queue away.  First odd queue in TX Bar */
4082 	nn->qcp_cfg = nn->tx_bar + NFP_QCP_QUEUE_ADDR_SZ;
4083 
4084 	/* Make sure the FW knows the netdev is supposed to be disabled here */
4085 	nn_writel(nn, NFP_NET_CFG_CTRL, 0);
4086 	nn_writeq(nn, NFP_NET_CFG_TXRS_ENABLE, 0);
4087 	nn_writeq(nn, NFP_NET_CFG_RXRS_ENABLE, 0);
4088 	err = nfp_net_reconfig(nn, NFP_NET_CFG_UPDATE_RING |
4089 				   NFP_NET_CFG_UPDATE_GEN);
4090 	if (err)
4091 		return err;
4092 
4093 	if (nn->dp.netdev) {
4094 		nfp_net_netdev_init(nn);
4095 
4096 		err = nfp_ccm_mbox_init(nn);
4097 		if (err)
4098 			return err;
4099 
4100 		err = nfp_net_tls_init(nn);
4101 		if (err)
4102 			goto err_clean_mbox;
4103 	}
4104 
4105 	nfp_net_vecs_init(nn);
4106 
4107 	if (!nn->dp.netdev)
4108 		return 0;
4109 	return register_netdev(nn->dp.netdev);
4110 
4111 err_clean_mbox:
4112 	nfp_ccm_mbox_clean(nn);
4113 	return err;
4114 }
4115 
4116 /**
4117  * nfp_net_clean() - Undo what nfp_net_init() did.
4118  * @nn:		NFP Net device structure
4119  */
4120 void nfp_net_clean(struct nfp_net *nn)
4121 {
4122 	if (!nn->dp.netdev)
4123 		return;
4124 
4125 	unregister_netdev(nn->dp.netdev);
4126 	nfp_ccm_mbox_clean(nn);
4127 	nfp_net_reconfig_wait_posted(nn);
4128 }
4129